Data Sheet 1 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Overview
The INFINEON Direct RDRAM is a general purpose high-performance memory device suitable for
use in a broad range of applications including computer memory, graphics, video, and any other
application where high bandwidth and low latency are required.
The 288-Mbit Direct Rambus DRAMs (RDRAM) are extremely high-speed CMOS DRAMs
organized as 16M words by 18 bits. The use of R ambus Signaling Level (RSL) technology permits
600 MHz to 800 MHz transfer rates while using conventional system and board design
technologies. Direct RDRAM devices are capable of sustained data transfers at 1.25 ns per two
bytes (10 ns per sixteen bytes).
The architecture of the Direct RDRAMs allows the highest sustained bandwidth for multiple,
simultaneous randomly addressed memory transacti ons. The separate control and data buses with
independent row and column control yield over 95% bus efficiency. The Direct RDRAM’s thirty-two
banks support up to four simultaneous t ransactions.
System oriented features for mobile, graphics and large memory systems include power
management, byte masking, and x18 organization. The two data bits in the x18 organization are
general and can be used for additional storage and bandwidth or for error correction.
Features
• Highest sustained bandwidth per DRAM devic e
– 1.6 GB/s sustained data transfer rate
– Separate control and data buses f or maximized efficiency
– Separate row and column control bus es for easy scheduling and highest perform ance
– 32 banks: four trans actions can take place simultaneously at full bandwidth data rates
• Low latency features
– Write buffer to reduce read lat ency
– 3 precharge mechanisms for controller flexibility
– Interleaved transactions
• Advanced power management:
– Multiple low power states allows flexibility in power consumption versus time to transition to
active state
– Power-down self-refresh
• Organization: 2 Kbyte pages and 32 banks, x18
– x18 organization allows ECC configurations or increased storage/bandwidth
• Uses Rambus Signaling Level (RSL) for up to 800 MHz operation
• The ODF function is allready implemented in this device and will be described in a lat er version
of this document.
Perliminary Datasheet
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 2 6.00
Figure 1 Direct RDRAM CSP Package
The 288-Mbit Direct RDRAMs are offered in a CSP horizontal package suitable for desktop as w ell
as low-profile add-in card and mobile applications.
Direct RDRAMs operate from a 2.5 V supply.
Table 1 Key Timing Parameters/Part Numbers
Organization I/O Freq.
MHz Tr ac Part Number
Normal Package:
16M × 18 600 53 ns HYB25R288180C-653
16M × 18 711 45 ns HYB25R288180C-745
16M × 18 800 45 ns HYB25R288180C-845
16M × 18 800 40 ns HYB25R288180C-840
Mirror Package:
16M × 18 600 53 ns HYB25M288180C-653
16M × 18 711 45 ns HYB25M288180C-745
16M × 18 800 45 ns HYB25M288180C-845
16M × 18 800 40 ns HYB25M288180C-840
R28818C-840
0010
Data Sheet 3 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Pinouts and Definitions
(9x8)
Center -Bonded Devices
These tables shows the pin assignments of the 9x8 center-bonded RDRAM package. The top
table is for the nor mal package, and bottom table is for the mirrored packa ge. The mec hanical
dimensions of this package are s hown in the last section.
Table 2: 9x8 Center-Bonded Device (top view for normal package)
10 NC NC
9NC DQA7 DQA4 CFM CFMN RQ5 RQ3 DQB0 DQB4 DQB7 NC
8GND VDD GND GNDa VDD GND VDD VDD GND
7NC CMD DQA5 DQA2 VDDa RQ6 RQ2 DQB1 DQB5 SIO1 NC
6
5
4NC SCK DQA6 DQA1 VREF RQ7 RQ1 DQB2 DQB6 SIO0 NC
3VCMOS GND VDD GND GND VDD GND GND VCMOS
2NC DQA8 DQA3 DQA0 CTMN CTM RQ4 RQ0 DQB3 DQB8 NC
1NC NC
Top view for
normal
package
ABCDEFGHJKLMN
Table 3: 9x8 Center-Bonded Device (top view for mirrored package)
10 NC NC
9NC DQA8 DQA3 DQA0 CTMN CTM RQ4 RQ0 DQB3 DQB8 NC
8VCMOS GND VDD GND GND VDD GND GND VCMOS
7NC SCK DQA6 DQA1 VREF RQ7 RQ1 DQB2 DQB6 SIO0 NC
6
5
4NC CMD DQA5 DQA2 VDDa RQ6 RQ2 DQB1 DQB5 SIO1 NC
3GND VDD GND GNDa VDD GND VDD VDD GND
2NC DQA7 DQA4 CFM CFMN RQ5 RQ3 DQB0 DQB4 DQB7 NC
1NC NC
Top view for
m irrored
package
ABCDEFGHJKLMN
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 4 6.00
Table 4
Signal I/O Type # Pins
Center Description
SIO1,SIO0 I/O CMOS1)
1) All CMOS signals are high-true; a high voltage is a logic one and a low voltage is logic zero.
2 Serial input/output. Pins for reading from and writing to
the control registers using a serial access protocol.
Also used for power management.
CMD I CMOS1) 1 Command input. Pins used in conjunction with SIO0
and SIO1 for reading from and writing to the control
registers. Also used for power management.
SCK I CMOS1) 1 Serial clock input. Clock source used for reading from
and writing to the control registers.
VDD –– 6 Supply voltage for the RDRAM core and interface logic.
VDDa –– 1 Supply voltage for the RDRAM analog circuitry.
VCMOS –– 2 Supply voltage for CMOS input/output pins.
GND –– 9 Ground reference for RDRAM core and interface.
GNDa –– 1 Ground reference for RDRAM analog circuitry.
DQA8 … DQA0 I/O RSL2)
2) All RSL signals are low-true; a low voltage is a logic one and a high voltage is logic zero.
9 Data byte A. Nine pins which carry a byte of read or
write data between the Channel and the RDRAM.
CFM I RSL2) 1 C lock from master. Interfac e clock used for receiving
RSL signals from the Channel. Positive polarity.
CFMN I RSL2) 1 Clock from master. Interface clock used for receiving
RSL signals from the Channel. Negative polarity
VREF 1 Logic threshold reference voltage for RSL signals
CTMN I RSL2) 1 Clock to master. Interface clock used for transmitting
RSL signals to the Channel. Negative polarity.
CTM I RSL2) 1 Clock to master. Interface clock used for transmitting
RSL signals to the Channel. Positive polarity.
RQ7 … RQ5 or
ROW2 … ROW0 IRSL
2) 3 Row access control. Three pins containing control and
address information for row accesses.
RQ4 … RQ0 or
COL4 … COL0 IRSL
2) 5 Column access control. Five pins containing control
and address information for column accesses.
DQB8 … DQB0 I/O RSL2) 9 Data byte B. Nine pins which carry a byte of read or
write data between the Channel and the RDRAM.
Total pin count per package 54 –
Data Sheet 5 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 2 288-MBit Direct RD RAM Block Diagram
Bank 31
DQA8..DQA0
1:8 Dem ux8:1 M ux
W rite Bu ffer
1:8 Dem ux
Write Bu ffer
8:1 Mux
Bank 30
Bank 29
Bank 18
Bank 17
Bank 16
Bank 15
Bank 14
Bank 13
Bank 1
Bank 0
SAmp
1/2
DQB8..DQB0
9
1:8 D em ux 1:8 Dem ux
P acket Decode
95
3
ROW2..ROW0 COL4..COL0CTM CTMN CFM CFMN
2
SCK,CMD
RCLKTCLK Control Registers
DCCOP CBC MAMBDXXOP BXDR RROP BR
887555556
95511
AV M S
Write
Buffer
MatchMatch
Mux
Match
DEVID
512x128x144
Internal DQB Data Path
Column Decode & Mask
72
9
9
72
9
DM
REFR
Row Decode
Mux
ACT
RD, WR
Power Modes
DRAM Core
Mux
XO P Decode
PREX
PREC
9 99
9
72
9
9 9
PRER
COLX COLC COLM
2
S IO0 ,S IO1
S ense Amp
Internal DQ A Data Path
P acket Decode
RO W A
RO WR
RCLK RCLK
RCLKTCLK
RCLK TCLK
R Q7 ..RQ5 o r R Q4 ..RQ0 o r
SAmp
0/1
SAmp
0
SAmp
14/15
SAmp
15
SAmp
13/14
SAmp
16/17
SAmp
17/18
SAmp
16
SAmp
29/30
SAmp
30/31
SAmp
31
64x72
SAmp
1/2
72
SAmp
0/1
SAmp
0
SAmp
14/15
SAmp
15
SAmp
13/14
SAmp
16/17
SAmp
17/18
SAmp
16
SAmp
29/30
SAmp
30/31
SAmp
31
64x72
64x72
Bank 2
•••
•••
•••
•••
•••
•••
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 6 6.00
General Description
Figure 2 is a bloc k diagram of the 288 Mbit Direct RDRAM. It consists of two major bloc ks: a “core”
block built from banks and sens e amps s imilar to those f ound in ot her t ypes of DRAM , and a D irect
Rambus interface block which permits an external controller to access this core at up to 1.6 GB/s.
Control Registers: The CMD, SCK, SIO0, and SIO1 pins appear in the upper center of Figure 2.
They are used to write and read a block of control registers. These registers supply the RDRAM
configuration information to a controller and they select the operating modes of the device. The nine
bit REFR value is used for tracking the last refreshed row. Most importantly, the five bit DEVID
specifies the device address of the RD RAM on the Channel.
Clocking: The CTM and CTMN pins (Clock-To-Master) generate TCLK (Transmit Clock), the
internal clock used to transmit read data. The CFM and CFMN pins (Clock-From-Master) generate
RCLK (Receive Clock), the internal clock s ignal us ed to receive w rite data and to receiv e the ROW
and COL pins.
DQA , DQB Pi ns: These 18 pins carry read (Q) and write (D) data across the Channel. They are
multiplexed/de-multiplexed from/to two 72-bit data paths (running at one-eighth the dat a frequency)
inside the RD RAM.
Banks: T he 32 Mbyte core of the RDRAM is divided into 32 one- Mbyte banks, each organized as
512 rows, with each row contai ni ng 128 dualocts (2 Kbytes), and each dualoct containing 16 bytes.
A dualoct is the smallest unit of data that can be addressed.
Sense Amps: The RDRAM contains 34 sense amps. Each sense amp consists of 1024 bytes of
fast storage (512 for DQA and 512 for DQB) and can hold one-half of one row of one bank of the
RDRAM. The s ense amp may hold any of the 512 half-rows of an ass ociated bank. However, each
sense amp is shared between two adjacent banks of the RDRAM (except for numbers 0, 15, 30, and
31). This introduces the restriction that adjacent banks may not be simultaneously acc essed.
RQ Pins: These pins carry control and address information. They are broken into two groups.
RQ7 … RQ5 are also called ROW2 … ROW0, and are used primarily for controlling row accesses.
RQ4 … RQ0 are also called COL4 … COL0, and are used primarily for controlling column
accesses.
ROW Pins: The principle use of these three pins is to manage the transfer of data between the
banks and the sense amps of the RDRAM. These pins are de-multiplexed into a 24-bit ROWA
(row-activate) or ROWR (row-operation) packet.
COL Pins: The principle use of these five pins is to manage the transfer of data between the
DQA/DQB pins and the sense amps of the RDRAM. These pins are de-multiplexed into a 23-bit
COLC (column-operation) packet and either a 17-bit COLM (mask) packet or a 17-bit COLX
(extended-operation) packet.
ACT Command: An ACT (activate) command from an ROWA packet causes one of the 512 rows
of the s elected bank t o be loaded to its associated s ense amps (t wo 512 byt e s ense amps f or D QA
and two for DQB).
PRER Command: A PRER (precharge) command from an ROWR packet causes the selected
bank to release its two associated sense amps, permitting a different row in that bank to be
activated, or permitting adjacent banks to be activated.
Data Sheet 7 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
RD Command: The RD (read) command causes one of the 128 dualocts of one of the sense amps
to be t ransmitted on the DQA/DQB pins of the Channel.
WR Command: The WR (w rite) command causes a dualoct received from the DQA/DQB data pins
of the Channel to be loaded into the write buffer. There is also space in the write buffer for the BC
bank address and C c olumn address information. The data in the write buffer is automatically retired
(written with optional bytemask) to one of the 128 dualocts of one of the sense amps during a
subsequent COP command. A ret ire can take place during a RD, WR, or NOCOP to another device,
or during a WR or NOCOP to the same device. T he write buffer will not retire during a RD to the
same device. The writ e buffer reduces the delay needed for the internal DQA/DQB data path turn-
around.
PREC Precharge: The PREC, RDA and WRA commands are similar to NOCOP, RD and WR,
except that a precharge operation is performed at the end of the column operation. These
commands provide a second mechanism for perf orming precharge.
PREX Precharge: After a RD command, or after a WR command with no byte masking (M = 0), a
COLX packet may be used to specify an extended operation (XOP). The most important XOP
command is PREX. This command provides a third mechanism for performing precharge.
Packet Format
Figure 3 shows the formats of the ROWA and ROWR packets on the ROW pins. Table 5 describes
the fields which comprise these packets. DR4T and DR4F bits are encoded to contain both the DR4
device address bit and a framing bit which allows the ROWA or R OWR packet to be recognized by
the RDRAM.
The AV (ROWA/ROWR packet selection) bit distinguishes between the two packet types. Both the
ROWA and R OWR pack et provide a five bit devic e addr ess and a five bit bank address. An ROWA
packet uses the remaining bits to specify a nine bit row address, and the ROWR packet uses the
remaining bits for an eleven bit opcode field. Note the use of the “RsvX” notation to reserve bits for
future address field extension.
Figure 3 also shows the formats of the COLC, COLM, and COLX packets on the COL pins. Table 6
describes the fields which comprise these packets.
Table 5 Field Description for ROW A Packet and ROWR Packet
Field Description
DR4T, DR4F Bits for framing (recognizing) a ROWA or ROWR packet. Also encodes
highest device address bit.
DR3 … DR0 Device address for ROWA or ROWR pack et.
BR4 … BR0 Bank address for ROWA or ROWR pack et. RsvB denotes bits ignored by the
RDRAM.
AV Selects between R OWA packet (AV = 1) and ROWR packet (AV = 0).
R8 … R0 Row address for ROW A packet. RsvR denotes bits ignored by the RD RAM.
ROP10 … ROP0 Opcode field for ROW R packet. Specifies precharge, refresh, and pow er
management functions.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 8 6.00
The COLC packet uses the S (Start) bit for framing. A C OLM or COLX packet is aligned with this
COLC packet, and is also framed by the S bit.
The 23 bit COLC packet has a five bit device address, a five bit bank address, a six bit column
address, and a four bit opcode. The COLC packet specifies a read or write command, as well as
some power management commands.
The remaining 17 bit s are interpreted as a COLM (M = 1) or COLX (M = 0) packet. A COLM packet
is used for a COLC write command which needs bytemask control. The COLM packet is associated
with the COLC packet from a time tRTR earlier. An COLX packet may be used to specify an
independent precharge command. It c ontains a five bit device address, a f ive bit bank address, and
a five bit opcode. The COLX pac ket may also be used to specify s ome housekeeping and power
management commands. The COLX packet is f ramed within a COLC packet but is not otherwise
associated with any other packet.
Table 6 Field Description for COLC Packet, COLM Packet, and COLX Packet
Field Description
S Bit for framing (rec ognizing) a COLC packet, and indir ec tly for framing COLM and
COLX packets.
DC4 … DC0 Device address for COLC packet.
BC4 … BC0 Bank address for COLC packet. RsvB denotes bits reserved for future extension
(controller drives 0’s).
C6 … C0 Column address for COLC packet. RsvC denotes bits ignored by the RDRAM.
COP3 …
COP0 Opcode field for COLC packet. Specifies read, write, precharge, and power
management functions.
M Selects between C OLM packet (M = 1) and COLX packet (M = 0).
MA7 … MA0 Bytemask write control bits. 1 = writ e, 0 = no-writ e. MA0 controls the earliest byte
on DQA8 … 0.
MB7 … MB0 Bytemask write control bits. 1 = writ e, 0 = no-writ e. MB0 controls the earliest byte
on DQB8 … 0.
DX4 … DX0 Device address for COLX packet.
BX4 … BX0 Bank address for COLX pack et. RsvB denotes bits reserved for fut ure extension
(controller drives 0’s).
XOP4 …
XOP0 Opcode field for COLX packet. Specifies precharge, IOL control, and power
management functions.
Data Sheet 9 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 3 Packet Formats
PREX (d0)
MSK (b1)
COLM Packet
T10
The COLM is associated with a previous COLC, and is aligned
with the present COLC, indicated by the Start bit (S = 1) position.
COL0
a)
COL1
MB5
MB6
MB2
MB3 MB0
COL2
COL3
COL4
CTM/CFM
MB7
M = 1 MA6
MB4 MB1
MA4 MA2
S=1
a
MA7 MA5 MA3
T8
COL0
DC0
COL1
COL2
DC1
DC2
COP3COP2
T9
COP0
COP1
The COLX is aligned with the present COLC,
indicated by the Start bit (S = 1) position.
COL0
b)
COL1
XOP0
XOP1
DX0
DX1 BX2
SPB04207
S=1
M = 0
MA0
MA1
COL2
COL3
CTM/CFM
COL4
C0BC0BC3
T11
BC4 BC1
RsvB BC2
C1
C2
T12
DQA8...0
DQB8...0
BX1
BX0
XOP2
XOP3
DX2
DX3
BX3
BX4
XOP4
DX4
b
RsvB
COLX Packet
T13 T14 T15
ROWA Packet
T2
COLC Packet
T2
T0
COL3
COL4
DC3
DC4
CTM/CFM
S = 1
T1
ROW1
ROW0
DR3
DR4F
ROW2
CTM/CFM
DR4T
BR2
DR0
DR1 BR1
AV=1
RsvB
BR4 RsvR
DR2 BR0 BR3
RsvR
T0 T1
DR2
DR0
DR1
C5
RsvC
C3
C4
T3
COL4...COL0
ROW2...ROW0
CTM/CFM
R3R6
R7 R4
R0
R1
R8 R5 R2
ROW1
ROW0
DR3
DR4F
CTM/CFM
ROW2
DR4T
PRER c0
t
Packet
WR b1
ACT a0
T2T0 T1 T3 T4 T7T5 T6 T8 T9 T14T12T10 T11 T13 T15
ROP8
ROP7
ROP6
RsvB
4RsvB
BR2
BR1
AV=0
ROP9
BR3BR0
ROP10
ROP1
ROP0
ROP3
ROP4
ROP2
ROP5
T3 T8
ROWR Packet
T9 T10 T11
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 10 6.00
Field Encoding Summary
Table 7 shows how the six device address bits are decoded for the ROWA and ROWR packets.
The DR4T and DR4F encoding merges a fifth device bit with a framing bit. When neither bit is
asserted, the device is not selected. Note that a broadc ast operation is indicated when both bits are
set. Broadcast operat ion w ould t ypically be us ed f or ref resh and power management commands. If
the device is s elected, the D M (DeviceMatch) signal is asserted and an AC T or R OP command is
performed.
Table 8 shows the enc odings of the remaining f ields of the ROWA and R OWR pac kets. An ROW A
packet is specified by asserting the AV bit. This causes the specified row of the specified bank of this
device to be loaded into the ass ociated sense amps.
An ROWR packet is specified w hen AV is not asserted. An 11 bit opc ode field encodes a command
for one of the banks of this device. The PRER command causes a bank and its two associated
sense amps to precharge, so another row or an adjacent bank may be activated. The REFA
(refresh-activate) command is similar to the ACT command, except the row address comes from an
internal register REFR, and REFR is incremented at the largest bank address. The REFP (refresh-
precharge) command is identical to a PRER command.
The NAPR, NAPRC, PDNR, ATTN, and RLXR commands are used for managing the power
dissipation of the RDRAM and are described in more detail in “Power State Management” on
page 58. The T CEN and T CAL commands are us ed t o adjus t the output driver slew rate and t hey
are described in more detail in “C urrent and Temperature Control” on page 65.
Table 7 Device Field Encodings for ROWA Packet and ROWR Packet
DR4T DR4F Device Selection Device Match Signal (DM )
1 1 All devices (broadcast) DM is set to 1
0 1 One device selected DM is set to 1 if {DEVID4 … DEVI D0} == {0, DR3 … DR0}
else DM is set to 0
1 0 One device selected DM is set to 1 if {DEVID4 … DEVI D0} == {1, DR3 … DR0}
else DM is set to 0
0 0 No packet present DM is s et to 0
Data Sheet 11 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Table 9 shows the COP field encoding. The device must be in t he ATTN power state in order to
receive COLC packets. The COLC packet is used primarily to specify RD (read) and WR (write)
commands. Retire operations (moving data from the write buffer to a sense amp) happen
automatically. See Figure 17 for a more detailed description.
The COLC packet c an also specify a PREC command, which prec harges a bank and it s associated
sense amps. The RDA/WRA commands are equiv alent to combining RD/WR with a PREC. RLXC
(relax) performs a power mode transition. See “Power State Management” on page 58.
Table 8 ROWA Packet and ROWR Packet Field En codings
DM1) AV ROP10…ROP0 Fi eld Name Comma nd Desc ription
1098765432:0
0 - ------------ No operation.
1 1 Row address ACT Activate row R8 … R0 of bank BR4 … BR0 of
device and move device to ATTN2).
10
11000x
3) x x 000 PRER P recharge bank BR4 … BR0 of this device.
10000
11 0 0 x 000 REFA Ref resh (activ ate) row REFR8 … REFR0 of bank
BR3 … BR0 of device.
Increment REFR if BR4 … BR0 = 1111 (see
Figure 50).
10
10101 0 0 x 000 R EFP Precharge bank BR4 … BR0 of this device after
REFA (see Figure 50).
1 0 xx00001 x 000 PDNR Move this device into the powerdown (PDN)
power state (see Figure 47).
1 0 xx00010 x 000 NAPR Move this device into the nap (NAP) po wer state
(see Figure 47).
1 0 xx00011 x 000 NAPRC M o ve this devi ce in to the na p (NAP) p ower state
conditionally.
1 0xxxxxxx0 000 ATTN2) Move this device into the attention (ATTN) power
state (see Figure 45).
1 0xxxxxxx1 000 RLXR Move this device into the standby (STBY) power
state (see Figure 46).
1 0 0000000x001 TCAL Temperature calibrate this device (see
Figure 52).
1 0 0000000x010 TCEN Temperature calibrate/enable this device (see
Figure 52).
1 0 00000000000NOROPNo operation.
1) The DM (Device Match signal) value is determined by the DR4T,DR4F, DR3 … DR0 field of the ROWA and
ROWR packets. See Table 7.
2) The ATTN c omma nd do es n ot c ause a R LX-t o-ATTN transi tion for a broadc ast oper ati on (DR4 T/DR4F = 1/1).
3) An “x” entry indicates which commands may be combined. For instance, the three commands
PRER/NAPRC/RLXR may be specified in one ROP value (011000111000).
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 12 6.00
Table 9 COLC Packet Field Encodings
SDC4 … DC0
(select device)1) COP3 … 0 Name Command Descr iption
0 ---- ----- – No operation.
1 /= (DEVID4 … 0) ----- – Retire wr ite buffer of this device.
1 == (DEVID4 … 0) x0002) NOCOP R etire write buffer of this device.
1 == (DEVID4 … 0) x001 WR Retire write buffer of this device, then write
column C5 … C0 of bank BC4 … BC0 to write
buffer.
1 == (DEVID4 … 0) x010 RSRV Reserved, no operation.
1 == (DEVID4 … 0) x011 RD Read column C 5 … C0 of bank BC4 … BC0 of
this device.
1 == (DEVID4 … 0) x100 PREC Retire writ e buffer of this device, then precharge
bank BC4 … BC0 (see Figure 14).
1 == (DEVID4 … 0) x101 WRA Same as WR, but precharge bank BC4 … BC0
after write buffer (with new data) is retired.
1 == (DEVID4 … 0) x110 RSRV Reserved, no operation.
1 == (DEVID4 … 0) x111 RDA Same as RD , but precharge bank BC4 … BC 0
afterward.
1 == (DEVID4 … 0) 1xxx RLXC Move this device into the standby (STBY) power
state (see Figure 46).
1) “/=” means not equal, “==” means equal.
2) An “x” entry indicates which commands may be combined. For instance, the two commands WR/R LXC may
be specified in one COP value (1001).
Data Sheet 13 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Table 10 shows the COLM and COLX field encodings. The M bit is asserted to specify a COLM
packet with two 8 bit bytemask fields MA and MB. If the M bit is not asserted, an COLX is specified.
It has device and bank address fields, and an opcode field. The primar y use of the COLX packet is
to permit an independent PREX (precharge) command to be specified without consuming control
bandwidth on the ROW pins. It is also used for the CAL (calibrate) and SAM (sample) current control
commands (see “Current and Temperature Control” on page 65), and for the RLXX power mode
command (see “Power State Management” on page 58).
Table 10 COLM Packet and COLX Packet Field Encodings
MDX4 … DX0
(selects device) XOP4 …
0Name Command Description
1 ---- – MSK MB/MA bytemasks used by WR/WRA.
0 /= (DEVID4 … 0) – – No operation.
0 == (DEVID4 … 0) 00000 NOXOP No operation.
0 == (DEVID4 … 0) 1xxx01)
1) An “x ” entry i ndica tes whi ch com mands ma y b e c ombin ed. For instance, th e two commands PREX/ RLX X may
be specified in one XOP value (10010).
PREX Precharge bank BX4 … BX0 of this device (see
Figure 14).
0 == (DEVID4 … 0) x10x0 CAL Calibrate (drive) IOL c urrent for this device (see
Figure 51).
0 == (DEVID4 … 0) x11x0 CAL/SAM Calibrate (drive) and Sample (update) IOL current
for this devic e (see Figure 51).
0 == (DEVID4 … 0) xxx10 RLXX Move this device into the standby (STBY) power
state (see Figure 46).
0 == (DEVID4 … 0) xxxx1 RSRV Reserved, no operation.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 14 6.00
DQ Packet T iming
Figure 4 shows the timing relationship of COLC packets with D and Q data packets. This document
uses a specific convention for measuring time intervals between packets: all packets on the ROW
and COL pins (ROWA, ROWR, COLC, COLM, COLX) use the trailing edge of the packet as a
reference point, and all packets on the DQA/DQB pins (D and Q) use the leading edge of the packet
as a ref erence point.
An RD or RDA command will transmit a dualoct of read data Q a time tCAC later. This time includes
one to five cycles of round-trip propagation delay on the Channel. The tCAC parameter may be
programmed to a one of a range of values (7, 8, 9, 10, 11, or 12 tCYCLE). The value chosen depends
upon the number of RDRAM devices on the Channel and t he RDRAM timing bin. See Figure 39 for
more information.
A WR or WRA comm and will receive a dualoct of write data D a time tCWD later. This time does not
need to include the round-trip propagation time of the Channel since the COLC and D packets are
traveling in the same direction.
When a Q packet follows a D packet (shown in t he left half of the figure), a gap (tCAC –tCWD) will
automatically appear between them because the tCWD value is always less than the tCAC value.
There will be no gap between the two COLC packets with the WR and RD commands which
schedule the D and Q packets.
When a D packet follows a Q packet (shown in the right half of the figure), no gap is needed between
them because the tCWD value is less than the tCAC v alue. However, a gap of tCAC –tCWD or greater
must be inserted bet ween the COLC packets with the RD WR commands by the cont roller so the
Q and D packets do not overlap.
Figure 4 Read (Q) and Write (D) Data Packet - Timing for tCAC = 7, 8, 9, 10, 11, or 12 tCYCLE
DQA8...0
DQB8...0
Q (y1)
CAC
t
Q (b1)
SPA04208
CAC
t
Q (c1) D (d1)
T25T5
RD b1
This gap on the DQA/DQB pins appears automatically
ROW0
ROW2...
COL4...COL0
WR a1
CTM/CFM
T0 T1 T2 T3 T4
-
CWD
t
t
CAC
t
CWD
T15T10T6 T7 T8 T9 T11 T12 T13 T14 T20T17T16 T18 T19 T22T21
This gap on the COL pins must be inserted by the controller
RD c1
CAC
t t
CWD
-
CWD
t
WR d1
T35T30T27T26 T28 T29 T32T31 T33 T34 T37T36 T38 T39 T42T41T40 T43 T44 T47T46T45
Data Sheet 15 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
COLM Packet to D Packet Mapping
Figure 5 show s a write operation initiated by a W R command in a COLC packet. If a subset of the
16 bytes of write data are to be written, then a COLM packet is transmitted on the COL pins a time
tRTR after the C OLC packet containing the WR command. The M bit of the COLM packet is set to
indicate that it c ontains t he M A and MB mask fields. Note t hat this COLM pac ket is aligned with the
COLC packet w hich causes the write buffer to be ret ired. See Figure 17 for more details.
If all 16 bytes of the D data packet are to be written, then no further control information is required.
The packet slot that would have been used by the COLM packet (tRTR after the COLC packet) is
available to be used as an COLX packet. This c ould be us ed for a PREX precharge c ommand or f or
a housekeeping command (this case is not shown). The M bit is not asserted in an COLX packet
and causes all 16 bytes of the previous WR to be written unconditionally. Note that a RD command
will never need a COLM packet, and will always be able to use the COLX packet option (a read
operation has no need for the byte-write-enable control bits).
Figure 5 also shows the mapping between the MA and MB fields of the C OLM packet and bytes of
the D packet on the DQA and DQB pins. Each mask bit controls whether a byte of data is written
(= 1) or not written (= 0).
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 16 6.00
Figure 5 Mapping Between COL M Packet and D Packet fo r WR Command
the M bit of the COLM packet is one.
(= 0) of the indicated DB bits when
controls writing (= 1) or no writing
Each bit of the MB7...MB0 field
the M bit of the COLM packet is one.
(= 0) of the indicated DA bits when
controls writing (= 1) or no writing
Each bit of the MA7...MA0 field
When M = 0, all data bytes are
When M = 1, the MA and MB
writing unconditionally.
individual data bytes.
fields control writing of
COL0
COL1
COL2
COL3
MB5
MB6
MB2
MB3 MB0
MB7
M = 1 MA6
MB4 MB1
MA4 MA2
DA55DA46DA37DA28DA19DA10
DA1 DA64
DQA1
MA1
DA9
DQA0 DA0
MA0 MA5
DA45
MA3
DA27
DA18
MA2
DA36
MA4
SPA04209
MA7
DA63
DA54
MA6
DB16
DB10
DB9
DA17
DA16
MB1
MB0
DQA7 DA7
DQA8 DA8
MA0
DQB0
DQB1
DB0
DB1
DQB7 DB7
MB7MB2 MB5
MB4
MB3 MB6
DA53
DA52DA34DA25 DA44
DA35DA26 DA45
DA70DA61
DA71
DA62
DB52
DB46
DB45
DB28
DB27
DB18
DB19
DB36
DB37
DB34
DB25 DB44
DB64
DB63
DB54
DB55
DB70
DB61
T12
T19
COLM Packet
a0 = {Da, Ba, Ra}
ACT a0
Transaction a: WR
T17
COL4
CTM/CFM
MA7 MA5 MA3
T18
DQA8...0
DQB8...0
COL4...COL0
ROW0
WR a1
T2
CTM/CFM
ROW2...
T1T0 T7T3 T4 T5 T6 T8 T9 T10 T11
ACT b0
PRER a2
DB17
a1 = {Da, Ba, Ca1}
MA1
T20
CTM/CFM
DQB8 DB8
T19
CWD
t
D (a1)
RTR
t
MSK (a1)
retire (a1)
a3 = {Da, Ba}
DB53DB35DB26 DB45
D Packet
T20 T21
DB71
DB62
T22
T17T14T13 T15 T16 T19T18 T20 T21 T24T23T22 T25 T26 T29T28T27 T30 T31 T34T33T32 T35 T36 T39T37 T38 T40 T41 T46T44T42 T43 T45 T47
Data Sheet 17 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
ROW-to-ROW Packet Interaction
Figure 6 ROW-to-ROW Packet Interaction-Timing
Figure 6 shows two packets on the ROW pins separated by an interval tRRDELAY which depends
upon the packet contents. No other ROW packets are sent to banks {Ba, Ba+1, Ba-1} between
packet “a” and packet “b” unless noted otherwise. Table 11 summarizes the tRRDELAY values for all
possible cases.
Cases RR1 through RR4 show two successive ACT commands. In case RR1, there is no restriction
since the ACT commands are to different devices. In case RR2, the tRR restriction applies to the
same device with non-adjacent banks . Cases RR3 and RR4 are illegal (as shown) since bank Ba
needs to be prec harged. If a PRER to Ba, Ba+1, or Ba-1 is inserted, tRRDELAY is tRC (tRAS to the PRER
command, and tRP to the next ACT).
Cases RR5 through RR8 show an ACT command followed by a PRER command. In cases RR5 and
RR6, there are no res trictions since the commands are to different devices or to non-adjacent banks
of the same device. In cases RR7 and RR8, the tRAS restriction means the activated bank m ust wait
before it can be precharged.
Cases RR9 through R R12 show a PR ER command followed by an ACT command. In c ases R R9
and RR10, there are essentially no restrictions s ince the commands are to different devices or to
non-adjacent banks of the same device. RR10a and RR10b depend upon whether a bracketed
bank (Ba ± 1) is precharged or activated. In cases RR11 and RR12, the same and adjac ent banks
must all w ait tRP for the sense amp and bank to precharge before being ac tivated.
Transaction b: ROPb
Transaction a: ROPa
DQB8...0
DQA8...0
b0 = {Db, Bb, Rb}
a0 = {Da, Ba, Ra}
SPT04210
COL4...COL0
ROW2...
CTM/CFM
ROPa a0
T0 T1 T3T2 T4 T17T11
ROPb b0
T8
RRDELAY
t
T5 T6 T7 T9 T10 T14T12 T13 T15 T16 T18 T19
ROW0
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 18 6.00
ROW-to-ROW Interaction (cont’d)
Cases RR13 through RR16 summarize the combinations of two successive PRER commands. In
case RR13 there is no restriction since two devices are addressed. In RR14, tPP applies , since the
same device is addressed. I n RR15 and RR16, the same bank or an adjacent bank m ay be given
repeated PRER commands with only the tPP restriction.
Two adjacent banks can’t be activate simultaneously. A precharge command to one bank will thus
affect the state of the adjacent banks (and sense amps). If bank Ba is activate and a PRER is
directed to Ba, then bank Ba will be precharged along with sense amps Ba-1/Ba and Ba/Ba+1. If
bank Ba+1 is activate and a PR ER is directed to Ba, t hen bank Ba+1 will be precharged along with
sense amps Ba/Ba+1 and Ba+ 1/Ba+2. If bank Ba-1 is activate and a PRER is directed to Ba, then
bank Ba-1 w ill be precharged along with sense amps Ba/Ba-1 and Ba-1/ Ba-2.
A ROW packet may contain commands other than ACT or PRER. The REFA and REFP commands
are equivalent to ACT and PRER for interaction analysis purposes. The interaction rules of the
NAPR, NAPRC, PDNR, RLXR, ATTN, TCAL, and TCEN commands are discussed in later sections
(see Table 8 for cross-ref).
Table 11 R OW-to-ROW Packet Interaction - Rules
Case # ROPa Da Ba Ra ROPb Db Bb Rb tRRDELAY Example
RR1 ACT Da Ba Ra ACT /= Da xxxx x…x tPACKET Figure 11
RR2 ACT Da Ba Ra ACT = = Da /= {Ba, Ba+1, Ba-1} x…x tRR Figure 11
RR3 ACT Da Ba Ra ACT == Da == {Ba+1, Ba-1} x …x tRC - illegal unless
PRER to Ba/Ba+1/Ba-1 Figure 10
RR4 ACT Da Ba Ra ACT == Da = = {Ba} x…x tRC - illegal unless
PRER to Ba/Ba+1/Ba-1 Figure 10
RR5 ACT Da Ba Ra PRER /= Da xxxx x…x tPACKET Figure 11
RR6 ACT Da Ba Ra PRER == Da /= {Ba, Ba+1, Ba-1 } x…x tPACKET Figure 11
RR7 ACT Da Ba Ra PRER == Da == {Ba+1, Ba-1} x…x tRAS Figure 10
RR8 ACT Da Ba Ra PRER == Da == {Ba} x…x tRAS Figure 15
RR9 PRER Da B a Ra ACT /= Da xxxx x…x tPACKET Figure 12
RR1 0 PRER Da Ba Ra A CT == Da /= {Ba, Ba±1,
Ba±2} x…x tPACKET Figure 12
RR10a PRER Da Ba Ra ACT == Da == {Ba+2 } x…x tPACKET/tRP if Ba+1 is
precharged/activated. –
RR10b PRER Da Ba Ra ACT == Da == {Ba-2} x…x tPACKET/tRP if Ba-1 is
precharged/activated. –
RR11 PRER Da Ba Ra ACT == Da == {Ba+1, Ba-1 } x…x tRP Figure 10
RR12 PRER Da Ba Ra ACT == Da == {Ba} x…x tRP Figure 10
RR13 PRER Da Ba Ra PRER /= Da xxxx x…x tPACKET Figure 12
RR1 4 PRER Da Ba Ra P RER == Da /= {Ba, Ba+1, Ba-1} x …x tPP Figure 12
RR15 PRER Da Ba Ra PRER == Da == {Ba+1, Ba-1} x…x tPP Figure 12
RR16 PRER Da Ba Ra PRER == Da == Ba x…x tPP Figure 12
Data Sheet 19 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
ROW-to-COL Packet Interaction
Figure 7 shows two packets on the ROW and COL pins. They must be separated by an interval
tRCDELAY which depends upon the pac ket contents. Table 12 summar izes the tRCDELAY values for all
possible cases. Note that if the COL packet is earlier than the ROW packet, it is considered a
COL-to-ROW packet interaction.
Cases RC1 through RC5 summarize the rules when the ROW packet has an ACT command.
Figure 15 and Figure 16 show examples of RC 5 - an activation followed by a read or writ e. RC4 is
an illegal situation, since a read or write of a precharged banks is being attempted (remember that
for a bank to be activated, adjacent banks must be precharged). In cases RC1, RC2, and RC3,
there is no interaction of the ROW and COL packets.
Figure 7 ROW-to-COL Packet Interaction - Timing
Cases RC6 through R C8 summarize the rules when the ROW packet has a PRER command. There
is either no interaction (RC6 t hrough RC9) or an illegal situation w ith a read or write of a precharged
bank (RC9).
The COL pins can als o s chedule a prec harge operat ion w ith a R DA, W RA, or PREC command in
a COLC packet or a PREX command in a COLX packet. The constraints of these precharge
operations may be converted to equivalent PRER command constraints using the rules
summarized in Figure 14.
Transaction b: COPb
Transaction a: ROPa
DQB8...0
DQA8...0
b1 = {Db, Bb, Cb1}
a0 = {Da, Ba, Ra}
SPT04211
COL4...COL0
ROW2...
CTM/CFM
COPb b1
T0 T1 T3T2 T4 T17T11
ROPa a0
T8
RCDELAY
t
T5 T6 T7 T9 T10 T14T12 T13 T15 T16 T18 T19
ROW0
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 20 6.00
COL-to-COL Packet I nteraction
Figure 8 COL-to-COL Packet I nteraction-Timing
Figure 8 shows three arbitrary pac kets on t he C OL pins . Packets “b” and “c” must be separated by
an interval tCCDELAY which depends upon the command and address values in all three packets.
Table 13 sum marizes the tCCDELAY values for all possible cases.
Cases CC1 through CC5 summarize t he rules for every situation other t han the case w hen COPb
is a WR command and COPc is a RD command. In CC3, when a RD command i s followed by a WR
command, a gap of tCAC –tCWD m ust be inserted between the two COL packets. See Figure 4 for
more explanation of why this gap is needed. For cases CC1, CC2, CC4, and CC5, there is no
restriction (tCCDELAY is tCC).
Table 12 R OW-to-COL Packet Interaction - Rules
Case # ROPa Da Ba Ra COPb Db Bb Cb1 tRCDELAY Example
RC 1 A CT Da Ba Ra NO CO P, RD, retire /= Da x xxx x…x 0 –
RC2 ACT Da Ba Ra NOCOP == Da xxxx x…x 0 –
RC3 ACT Da Ba Ra RD, retire == Da /= {Ba, Ba+1, Ba-1} x…x 0 –
RC4 A CT Da Ba Ra RD, retire == Da == {Ba+1, Ba-1} x… x Illegal –
RC5 ACT Da Ba Ra RD, retire == Da == Ba x…x tRCD Figure 15
RC 6 P RE R Da Ba Ra N OCOP, RD, retire /= Da x xxx x… x 0 –
RC7 PRER Da Ba Ra NOCOP == Da xxxx x…x 0 –
RC8 PRER Da Ba Ra RD, retire == Da /= {Ba, Ba+1, Ba-1} x…x 0 –
RC9 P RER Da Ba Ra R D, retire == Da == {Ba+1, B a-1} x…x Illegal –
Transaction b: COPb
Transaction a: COPa
DQB8...0
DQA8...0
b1 = {Db, Bb, Cb1}
a1 = {Da, Ba, Ca1}
SPT04212
COL4...COL0
ROW2...
CTM/CFM
T0 T1 T3T2 T4 T17T11T8
CCDELAY
t
T5 T6 T7 T9 T10 T14T12 T13 T15 T16 T18 T19
ROW0
Transaction c: COPc c1 = {Dc, Bc, Cc1}
COPa a1 COPb b1 COPc c1
Data Sheet 21 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
In cases CC6 through CC10, COPb is a WR command and COPc is a RD command. The tCCDELAY
value needed between these two packets depends upon t he c ommand and addres s in the packet
with COPa. In particular, in case C C6 when there is WR-WR-RD command sequence directed to
the same device, a gap will be needed between the packets with COPb a nd COPc. The gap will
need a COLC packet with a NOCOP command directed to any device in order to force an automatic
retire to tak e place. Figure 18 (right) provides a more detailed explanation of this case.
In case CC10, there is a RD -WR-RD sequence directed to the same device. If a prior write to the
same device is unretired when COPa is issued, then a gap will be needed between the packets with
COPb and COPc as in case CC6. The gap will need a COLC packet with a NOCOP command
directed to any device in order t o force an automatic retire to tak e place.
Cases CC7, CC 8, and CC9 have no res triction (tCCDELAY is tCC).
For the purposes of analyzing C OL-to-ROW int eractions, t he PR EC , WRA, and RD A c ommands of
the COLC packet are equivalent to the NOCOP, WR, and RD commands. These commands also
cause a precharge operation PREC to take place. This precharge may be converted to an
equivalent PRER com mand on the ROW pins using the rules summarized in Figure 14.
Table 13 C OL-to-COL Packet Interaction - Rules
Case # COPa Da Ba Ca1 COPb Db Bb Cb1 COPc Dc Bc Cc1 tCCDELAY Example
CC1 xxxx xxxxx x…x x…x NOCOP Db Bb Cb1 xxxx xxxxx x…x x…x tCC –
CC2 xxxx xxxxx x…x x…x RD,WR Db Bb Cb1 NOCOP xxxxx x…x x…x tCC –
CC3 xxxx xxxxx x…x x…x RD Db Bb Cb1 WR xxxxx x…x x…x tCC+tCAC-tCWD Figure 4
CC4 xxxx xxxxx x…x x…x RD Db Bb Cb1 RD xxxxx x…x x…x tCC Figure 15
CC5 xxxx xxxxx x…x x…x WR Db Bb Cb1 WR xxxxx x…x x…x tCC Figure 16
CC 6 WR == Db x x… x W R D b Bb Cb 1 RD = = Db x… x x… x tRTR Figure 18
CC 7 WR == Db x x… x W R D b Bb Cb 1 RD /= Db x… x x… x tCC –
CC8 WR /= Db x x…x WR Db Bb Cb1 RD == Db x…x x…x tCC –
CC9 NOC OP == Db x x…x WR D b Bb Cb1 RD = = Db x… x x…x tCC –
CC 10 RD == Db x x…x WR D b Bb Cb 1 RD = = Db x… x x… x tCC –
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 22 6.00
COL-to-ROW Packet I nteraction
Figure 9 COL-to-ROW Packet I nteraction - Timing
Figure 9 shows arbitrary packets on the COL and ROW pins. They must be separated by an
interval tCRDELAY which depends upon the command and address values in the packets. Table 14
summarizes the tCRDELAY value for all possible cases.
Cases CR1, CR2, CR3, and CR9 show no interaction between the COL and R OW packets, either
because one of the commands is a NOP or because the packets are directed to different devices or
to non-adjacent bank s.
Case CR4 is illegal because an already-activated bank is to be re-activated without being
precharged Case CR5 is illegal bec ause an adjacent bank can’t be activated or precharged until
bank Ba is precharged first.
In case CR6, the COLC packet contains a RD command, and the ROW packet contains a PRER
command for the same bank. The tRDP parameter specifies the required spacing.
Likewise, in case CR7, the COLC packet causes an automatic retire to take place, and the ROW
packet contains a PRER command for the s ame bank. The tRTP parameter specifies the required
spacing.
Case CR8 is labeled “Hazardous” because a WR command should always be followed by an
automatic retire before a precharge is scheduled. Figure 19 shows an example of what can happen
when the retire is not able to happen before the precharge.
For the purposes of analyzing C OL-to-ROW int eractions, t he PR EC , WRA, and RD A c ommands of
the COLC packet are equivalent to the NOCOP, WR, and RD commands. These commands also
cause a precharge operation to take place. This precharge may converted to an equivalent PRER
command on t he ROW pins using the rules summarized in Figure 14.
A ROW packet may contain commands other than ACT or PRER. The REFA and REFP commands
are equivalent to ACT and PRER for interaction analysis purposes. The interaction rules of the
NAPR, PDNR, and RLXR commands are discussed in a later section.
Transaction b: ROPb
Transaction a: COPa
DQB8...0
DQA8...0
b0 = {Db, Bb, Rb}
a1 = {Da, Ba, Ca1}
SPT04213
COL4...COL0
ROW2...
CTM/CFM
COPa a1
T0 T1 T3T2 T4 T17T11
ROPb b0
T8
CRDELAY
t
T5 T6 T7 T9 T10 T14T12 T13 T15 T16 T18 T19
ROW0
Data Sheet 23 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Table 14 C OL-to-ROW Packet Interaction - Rules
Case # COPa Da Ba Ca1 ROPb D b Bb Rb tCRDELAY Example
CR1 NOCOP Da Ba Ca1 x…x xxxxx xxxx x…x 0 –
CR 2 RD /WR Da Ba Ca1 x… x /= Da x xxx x…x 0 –
CR3 RD/WR Da Ba Ca1 x…x == Da /= {Ba, Ba+1, Ba-1} x…x 0 –
CR 4 RD /WR Da Ba Ca1 A CT == Da == {Ba} x… x I llega l –
CR5 RD/WR Da Ba C a1 ACT == Da == {Ba+1, Ba-1} x…x Illegal –
CR6 RD D a Ba Ca1 PRER == Da == {Ba, Ba+ 1, Ba-1} x…x tRDP Figure 15
CR7 retire1) Da Ba Ca1 PRER == Da == {Ba, Ba+1, Ba-1} x…x tRTP Figure 16
CR8 WR2) Da Ba Ca1 PRER == Da == {Ba, Ba+1, Ba-1} x…x 0 Figure 19
CR9 xxxx Da Ba Ca1 NOROP xxxxx xxxx x…x 0 –
1) This is any command which permits the write buffer of device Da to retire (see Table 9). “Ba” is the bank address in the wr ite
buffer.
2) This situati on is hazardous because the write buffer will be left unretired while the targeted bank is precharged. See Figure 19.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 24 6.00
ROW-to-ROW Examples
Figure 10 shows examples of some of the ROW-to-ROW packet spacings from Table 11. A
complete sequence of activate and precharge commands is directed to a bank. The RR8 and RR12
rules apply to this sequence. In addition to satisfying the tRAS and tRP timing parameters, the
separation between ACT commands to the same bank must also satisfy the tRC timing parameter
(RR4).
When a bank is activated, it is necessary for adjac ent banks to remain precharged. As a result, the
adjacent banks will also satisfy parallel timing constraints; in the example, the RR11 and RR3 rules
are analogous to the RR12 and RR 4 rules.
Figure 10 Row Packet Example
Figure 11 shows examples of the ACT-to-ACT (RR1, RR2) and ACT-to-PRER (RR5, RR6)
command spacings from Table 11. In general, the commands in ROW packets may be spaced an
interval tPACKET apart unless they are directed to the same or adjacent banks or unless they are a
similar command t ype (both PRER or both ACT) directed to the same device.
DQA8...0
DQB8...0
COL4...COL0
t
RC
SPA04214
T24
Same Device
Same Device
Same Device
Same Device
Same Device
ROW0
CTM/CFM
ROW2...
ACT a0
T2T0 T1 T3 T4 T14T7T5 T6 T8 T9 T10 T11 T13T12
RAS
t
T19T15 T16 T18T17 T20 T21 T23T22
a0 = {Da, Ba, Ra}
a1 = {Da, Ba+1}
b0 = {Da, Ba+1, Rb}
b0 = {Da, Ba, Rb}
b0 = {Da, Ba+1, Rb}
b0 = {Da, Ba, Rb}
RR11
Adjacent Bank
T34
RR12
ACT b0
PRER a1
RP
t
T29
Same Bank
T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43
RR7
RR4
RR3
Adjacent Bank
Adjacent Bank
Same Bank
T46T45T44 T47
Data Sheet 25 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 11 Row Packet Example
Figure 12 s hows examples of the PRER-to-PRER (RR13, RR14) and PRER -to-ACT (RR9, RR10)
command spacings from Table 12. The RR15 and RR16 cases (PRER-to-PRER to same or
adjacent banks) are not shown, but are similar to RR14. In general, the commands in ROW packets
may be spaced an interval tPACKET apart unless they are directed to the sam e or adjacent banks or
unless they are a similar command type (both PRER or both ACT) directed to the same device.
DQA8...0
DQB8...0
COL4...COL0
SPA04215
T24
Different Device
Same Device
Same Device
Different Device
ACT b0
ROW0
CTM/CFM
ROW2...
ACT a0
PACKET
t
T2T0 T1 T3 T4 T14
ACT a0
T7T5 T6 T8 T11 T13T12
t
ACT c0
RR
T19T15 T16 T18T17 T20 T21 T23T22
a0 = {Da, Ba, Ra}
b0 = {Db, Bb, Rb}
c0 = {Da, Bc, Rc}
b0 = {Db, Bb, Rb}
c0 = {Da, Bc, Rc}
Any Bank
T34
Non-adjacent Bank
PRER b0
PACKET
ACT a0
t
T29T27 T28 T30 T31 T32 T33
ACT a0
PACKET
t
T39T36T35
RR6
T41T40 T42 T43
Non-adjacent Bank
Any Bank RR1
RR5
RR2
PRER c0
T46T45T44 T47
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 26 6.00
Figure 12 Row Packet Examples
Row and Co lumn Cycle Description
Activate: A row cycle begins with the activate (ACT) operation. The activation process is
destructive; the act of sensing the value of a bit in a bank’s storage cell transfers the bit to the sense
amp, but leaves the original bit in the storage cell with an incorrect value.
Restore: Because the activation process is destructive, a hidden operation called restore is
automatically performed. The restore operation rewrites the bits in the sense amp back into the
storage cells of the activated row of the bank.
Read/Write: While the res tore operation takes place, the sense amp m ay be read (RD) and written
(WR) using column operations. If new data is written into the sense amp, it is automatically
forwarded to the storage cells of the bank so the data in the activated row and t he data in the sense
amp remain identic al.
Precharge: When both the restore operation and the column operations are completed, the sense
amp and bank are precharged (PRE). This leaves them in the pr oper state to begin another activate
operation.
Intervals: The activate operation requires the interval tRCD,MIN to complete. The hidden restore
operation requires the interval tRAS,MIN –tRCD,MIN to complete. Column read and write operations are
also performed during the tRAS,MIN –tRCD,MIN interval (if more than about four column operations are
COL4...COL0
DQB8...0
DQA8...0
SPA04216
T24
Same Device
Same Device
Same Device
Different Device
Same Device
Different Device
PRER b0
ROW2...
CTM/CFM
ROW0
PRER a0
PACKET
t
T2T0 T1 T3 T4 T14
PRER a0
T7T5 T6 T8 T11 T13T12
t
PRER c0
PP
T19T15 T16 T18T17 T20 T21 T23T22
c0 = {Da, Ba+1, Rc}
b0 = {Db, Bb, Rb}
c0 = {Da, Bc, Rc}
c0 = {Da, Ba, Rc}
c0 = {Da, Bc, Rc}
b0 = {Db, Bb, Rb}
a0 = {Da, Ba, Ra}
Any Bank
T34
Non-adjacent Bank
ACT b0
PRER a0
PACKET
t
T29T27 T28 T30 T31 T32 T33
PRER a0
PACKET
t
T39T36T35
RR10
T41T40 T42 T43
Non-adjacent Bank
Any Bank
Same Bank
Adjacent Bank
RR13
RR15
RR16
RR9
RR14
ACT c0
T46T45T44 T47
Data Sheet 27 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
performed, this interval must be increased). The precharge operat ion requires the interval tRP,MIN to
complete.
Adjacent Banks: An RDRAM with a “s” designation (256K ×32s ×16/18) indicates it contains “split
banks”. This m eans the sense amps are s hared between two adjacent banks . The only exception
is that sense amp 0, 15, 30, and 31 are not shared. When a row in a bank is activated, the two
adjacent sense amps are connected to (associated with) that bank and are not available for use by
the two adjacent banks. These two adjacent banks must remain precharged while the selected bank
goes through its activate, restore, read/write, and precharge operations.
For example (referring to the block diagram of Figure 2), if bank 5 is accessed, sense amp 4/5 and
sense amp 5/6 will both be loaded with one of the 512 rows (with 512 bytes loaded into eac h sense
amp from the 1 Kbyte row - 256 bytes to t he DQA side and 256 bytes to the DQB side). While this
row from bank 5 is being ac cessed, no rows m ay be accessed in banks 4 or 6 bec ause of the sense
amp sharing.
Precharge Mechanisms
Figure 13 shows an example of precharge with the ROWR packet mechanism. The PRER
command must occur a time tRAS after the ACT command, and a time tRP before the next ACT
command. This timing will serve as a baseline against which the other precharge mechanisms can
be compared.
Figure 13 Precharge via PRER Command in ROWR Packet
Figure 14 (t op) s hows an example of precharge with a R DA command. A bank is ac tivated with an
ROWA packet on the ROW pins. Then, a series of four dualocts are read with RD commands in
COLC packets on the COL pins. The fourth of these commands is a RDA, which causes t he bank
to automatically prec harge when the final read has finished. The timing of this automatic prec harge
is equivalent to a PRER command in an ROWR packet on the ROW pins that is offset a time tOFFP
SPA04217
T24
ROW2...
COL4...COL0
DQB8...0
DQA8...0
RP
t
CTM/CFM
T2T1T0 T3 T4 T14T7T5 T6 T8 T9 T12T10 T11 T13 T19T17T15 T16 T18 T20 T21 T23T22 T44T34T29T25 T26 T28T27 T30 T31 T33T32 T39T35 T36 T37 T38 T40 T41 T42 T43 T46T45 T47
ACT a0 ACT b0
PRER a5
ROW0
RAS
t
t
RC
a0 = {Da, Ba, Ra}
a5 = {Da, Ba}
b0 = {Da, Ba, Rb}
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 28 6.00
from the COLC packet with the RDA command. The RDA command should be treated as a RD
command in a COLC packet as well as a s imultaneous (but offset) PRER command in an ROW R
packet when analyz ing interactions with other packets.
Figure 14 (middle) shows an example of precharge with a WRA command. As in the RDA example,
a bank is activated with an ROWA packet on the ROW pins. Then, two dualocts are written with WR
commands in COLC pac kets on the C OL pins. The second of these commands is a WRA, w hich
causes the bank to automatically precharge when the final write has been retired. The timing of this
automatic precharge is equiv alent to a PRER command in an ROWR packet on the ROW pins that
is offset a t ime tOFFP from the COLC packet that causes the automatic retire. The WRA command
should be treated as a WR command in a COLC packet as well as a simultaneous (but offset) PRER
command in an ROWR packet when analyzing interactions with other packets. Note that the
automatic retire is triggered by a COLC packet a time tRTR after the COLC packet with the WR
command unless the second COLC contains a RD command to the same device. This is described
in more detail in Figure 17.
Figure 14 (bottom) shows an example of precharge with a PREX command in an COLX packet. A
bank is activated with an ROWA packet on the ROW pins. Then, a series of four dualocts are read
with RD commands in COLC packets on the COL pins. The fourth of these COLC packets includes
an COLX packet with a PREX command. This causes the bank to precharge with timing equivalent
to a PRER command in an ROWR packet on the ROW pins that is offset a time tOFFP from the COLX
packet with the PREX command.
Data Sheet 29 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 14 Offsets for Alternate Precharge Mechanisms
D (a2)
a1 = {Da, Ba, Ca1}
retire (a2)
MSK (a2)
T24
PREX a5
a1 = {Da, Ba, Ca1}
a3 = {Da, Ba, Ca3}
Transaction a: RD
DQB8...0
DQA8...0
a0 = {Da, Ba, Ra}
Q (a1)
Transaction a: WR
COLC Packet: PREX Precharge Offset
The PREX precharge is equivalent to a PRER command here
CTM/CFM
ROW0
COL4...COL0
ROW2...
ACT a0
T2T0 T1 T3 T4
COL4...COL0
DQB8...0
DQA8...0
T14
RD a2RD a1
T7T5 T6 T8 T9 T10 T11 T13T12
RD a3 RD a4
T19T15 T16 T18T17 T20 T21 T23T22
a0 = {Da, Ba, Ra}
D (a1)
retire (a1)
MSK (a1)
RTR
t
a5 = {Da, Ba}
Q (a4)
a2 = {Da, Ba, Ca2}
a4 = {Da, Ba, Ca4}
Q (a2) Q (a3)
SPA04218
a5 = {Da, Ba}
T34
ACT b0
PRER a5
OFFP
t
T29T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43
a2 = {Da, Ba, Ca2}
OFFP
t
T46T45T44 T47
T24
T24
a1 = {Da, Ba, Ca1}
a3 = {Da, Ba, Ca3}
The WRA precharge (triggered by the automatic retire) is equivalent to a PRER command here
Transaction a: RD
COLC Packet: WRA Precharge Offset
The RDA precharge is equivalent to a PRER command here
ROW2...
ROW0
CTM/CFM
ACT a0
T2T0 T1 T3 T4
ROW2...
ROW0
DQB8...0
DQA8...0
COL4...COL0
ACT a0
T14
a0 = {Da, Ba, Ra}
T7T5 T6 T8 T9 T10 T11 T13T12 T19T15 T16 T18T17 T20 T21 T23T22
RD a2RD a1 RD a3 RD a4
Q (a1)
COLC Packet: RDA Precharge Offset
CTM/CFM
T2T0 T1 T3 T4 T14T7T5 T6 T8 T9 T10 T11 T13T12 T19T15 T16 T18T17 T20 T21 T23T22
a5 = {Da, Ba}
T34
ACT b0
a2 = {Da, Ba, Ca2}
a4 = {Da, Ba, Ca4}
PRER a5
T29T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43
ACT b0
Q (a4)Q (a2) Q (a3)
PRER a5
OFFP
t
T46T45T44 T47
T34T29T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43 T46T45T44 T47
WR a1 WRA a2
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 30 6.00
Read Transaction - Example
Figure 15 shows an example of a read t ransaction. It begins by activating a bank with an ACT a0
command in an ROWA packet. A time tRCD later a RD a1 command is issued in a COLC packet.
Note that the ACT c ommand inc ludes the device, bank , and row addres s (abbreviated as a0) w hile
the RD command includes device, bank, and column address (abbreviated as a1). A time tCAC after
the RD command the read data dualoct Q(a1) is returned by the device. Note that the packets on
the ROW and C OL pins us e t he end of t he pac ket as a timing reference point, while t he pac kets on
the DQA/DQB pins use the beginning of the packet as a timing reference point.
A time tCC after the first COLC packet on the COL pins a second is issued. It contains a RD a2
command. The a2 address has the same device and bank address as the a1 address (and a0
address), but a different column address. A time tCAC after the second RD command a second read
data dualoct Q(a2) is returned by the device.
Next, a PRER a3 command is issued in an ROWR packet on the ROW pins. This causes the bank
to precharge so that a different row may be activated in a subsequent transaction or so that an
adjacent bank may be activated. The a3 address includes the same device and bank address as the
a0, a1, and a2 addres ses. The PRER command must occur a t ime tRAS or more after the original
ACT command (the activation operation in any DRAM is destructive, and the contents of the
selected row must be restored from the two associated sense amps of the bank during the tRAS
interval). The PRER c ommand must also occur a time tRDP or more after the last RD command. Note
that the tRDP value shown is greater than the tRDP,MIN specification in Table 23. This transaction
example reads two dualocts, but there is actually enough time to read three dualocts before tRDP
becomes the limiting parameter rather t han tRAS. If four dualocts were read, the packet with PRER
would need to shift right (be delay ed) by one tCYCLE (note - this case is not shown).
Finally, an ACT b0 command is issued in an ROWR packet on the ROW pins. The second ACT
command must occur a time tRC or more after the first ACT command and a time tRP or more after
the PRER command. This ensures that the bank and its associated sense amps are prec harged.
This example assumes that the s econd transaction has the same device and bank address as the
first transaction, but a different row address. Transaction b may not be started until transaction a has
finished. However, transactions to other banks or other devices may be issued during transaction a
Data Sheet 31 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 15 Read Transaction Example
Write Transaction - Example
Figure 16 shows an example of a write transaction. It begins by activating a bank with an ACT a0
command in an ROWA packet. A time tRCD –tRTR later a WR a1 command is issued in a COLC
packet (note that the tRCD interval is measured to the end of the COLC packet with the first retire
command). Note that the ACT command includes the device, bank, and row address (abbreviated
as a0) while the WR command includes device, bank, and column address (abbreviated as a1). A
time tCWD after the WR command the write data dualoct D(a1) is issued. Note that the packets on the
ROW and C OL pins us e t he end of t he pac ket as a timing reference point, while t he pac kets on the
DQA/DQB pins use the beginning of the packet as a timing reference point.
A time tCC after the first COLC packet on the COL pins a second COLC packet is issued. It contains
a WR a2 command. The a2 address has the same device and bank address as the a1 address (and
a0 address), but a diff erent column address. A time tCWD after the second WR command a second
write data dualoc t D(a2) is issued.
A time tRTR after each WR command an optional COLM packet MSK (a1) is issued, and at the same
time a COLC packet is issued causing the write buffer to automatically retire. See Figure 17 for
more detail on the write/retire mechanism. If a COLM packet is not used, all data bytes are
unconditionally written. If the COLC pac ket which c auses the write buffer t o retire is delayed, then
the COLM packet (if used) must also be delayed.
Next, a PRER a3 command is issued in an ROWR packet on the ROW pins. This causes the bank
to precharge so that a different row may be activated in a subsequent transaction or so that an
adjacent bank may be activated. The a3 address includes the same device and bank address as the
a0, a1, and a2 addres ses. The PRER command must occur a t ime tRAS or more after the original
a1 = {Da, Ba, Ca1}
Transaction b: XX
Transaction a: RD
DQA8...0
DQB8...0
t
b0 = {Da, Ba, Rb}
a0 = {Da, Ba, Ra}
RCD
ttCAC
CC
Q (a1)
t
CAC
t
RDP
a3 = {Da, Ba}a2 = {Da, Ba, Ca2}
Q (a2)
SPT04219
T24
ROW2...
CTM/CFM
ROW0
COL4...COL0
ACT a0
T2T0 T1 T3 T4
RD a2RD a1
RAS
t
tRC
T14T7T5 T6 T8 T9 T10 T11 T13T12 T19T15 T16 T18T17 T20 T21 T23T22
ACT b0
PRER a3
T34T29T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43 T46T45T44 T47
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 32 6.00
ACT command (the activation operation in any DRAM is destructive, and the contents of the
selected row must be restored from the two associated sense amps of the bank during the tRAS
interval).
A PRER a3 command is issued in an ROWR packet on the ROW pins. The PRER command must
occur a tim e tRTP or more after the last COLC which causes an automatic retire.
Finally, an ACT b0 command is issued in an ROWR packet on the ROW pins. The second ACT
command must occur a time tRC or more after the first ACT command and a time tRP or more after
the PRER command. This ensures that the bank and its associated sense amps are prec harged.
This example assumes that the s econd transaction has the same device and bank address as the
first transaction, but a different row address. Transaction b may not be started until transaction a has
finished. However, transactions to other banks or other devices may be issued during transaction a.
Figure 16 Write Transaction Example
D (a2)
MSK (a2)
retire (a2)
a1 = {Da, Ba, Ca1}Transaction a: WR
Transaction b: XX
DQB8...0
DQA8...0
COL4...COL0
t
WR a2
a0 = {Da, Ba, Ra}
b0 = {Da, Ba, Rb}
WR a1
D (a1)
CWD
retire (a1)
MSK (a1)
t
t
CWD
CC
a3 = {Da, Ba}a2 = {Da, Ba, Ca2}
SPT04220
T24
CTM/CFM
ROW2...
ROW0
ACT a0
T2T0 T1 T3 T4
RCD
tt
RTR
t
RTR
t
RAS
t
RC
T14T7T5 T6 T8 T9 T10 T11 T13T12 T19T15 T16 T18T17 T20 T21 T23T22
ACT b0
PRER a3
RTP
t
RP
t
T34T29T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43 T46T45T44 T47
Data Sheet 33 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Write/Retire - Examp les
The process of writing a dualoct into a sense amp of an RDRAM bank occurs in two steps. The first
step consists of t ransporting t he w rite command, w rite address, and write data into the write buf fer.
The second step happens when the RDRAM automatically retires the write buffer (with an optional
bytemask) into the s ense amp. This two-step write process reduces the natural turn-around delay
due to the internal bidirectional data pins .
Figure 17 (left) shows an example of this two step process. The first COLC packet contains the WR
command and an address specifying device, bank and column. The write data dualoct follows a
time tCWD later. This information is loaded int o the write buffer of the s pecified device. The C OLC
packet which follows a time tRTR later will retire the write buffer. The ret ire w ill happen automatically
unless (1) a COLC packet is not framed (no COLC packet is present and the S bit is zero), or (2) the
COLC packet contains a RD command to the same device. If the retire does not take place at time
tRTR after the original WR command, t hen the device c ontinues to frame C OLC packets, looking f or
the first that is not a R D directed to i tself. A bytemask MSK(a1) may be supplied in a COLM pac ket
aligned with the COLC that retires the write buffer at time tRTR after the WR command.
The memory controller must be aware of this two-step write/retire process. Controller performance
can be improved, but only if the controller design accounts for several side ef fects.
Figure 17 Normal Retire (left) and Retire/Read Ordering (right)
Figure 17 (right) shows the first of these side effects. T he f irst COLC pac ket has a WR command
which loads the address and data into the write buffer. The third COLC causes an automatic retire
of the write buf fer to the sense amp. The second and fourth COLC packets (which bracket the ret ire
packet) contain RD commands with the same dev ice, bank and column address as the original WR
command. In other words, the same dualoct address that is written is read both before and after it
is actually retired. The first RD returns the old dualoct value from the sense amp before it is
overwritten. The sec ond RD returns the new dualoct value that was just written.
Transaction b: RD
Transaction c: RD
Transaction a: WR
DQA8...0
DQB8...0
Transaction a: WR a1 = {Da, Ba, Ca1}
CWD
tt
RTR
D (a1)
DQA8...0
DQB8...0
D (a1)
c1 = {Da, Ba, Ca1}
b1 = {Da, Ba, Ca1}
a1 = {Da, Ba, Ca1}
CWD
t
RTR
t
Q (b1)
SPT04221
Q (
WR a1
This RD gets the old data
Retire is automatic here unless:
(1) No COLC packet (S = 0) or
(2) COLC packet is RD to device Da
CTM/CFM
COL0
COL4...
ROW0
ROW2...
WR a1
T2T0 T1 T3 T4
retire (a1)
MSK (a1)
COL0
COL4...
ROW0
ROW2...
CTM/CFM
T7T5 T6 T8 T9 T12T10 T11 T13 T14 T0 T1
This RD gets the new data
CAC
RD b1 MSK (a1)
retire (a1)
t
RD c1
CAC
t
T4T2 T3 T5 T6 T9T7 T8 T10 T11 T14T13T12 T15 T16 T19T18T17 T20 T21 T23T22
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 34 6.00
Figure 18 (left) shows the result of performing a RD command to the same device in the same
COLC packet slot that would normally be used for the retire operation. The read may be to any bank
and column address; all that matters is t hat it is to the same device as the WR command. The retire
operation and MSK(a1) will be delayed by a t ime tPACKET as a result. If the R D command us ed the
same bank and column address as the WR command, t he old dat a f rom the s ense amp would be
returned. If many RD com mands to the same device were issued instead of the single one that is
shown, then the retire operation would be held off an arbitrarily long time. However, once a R D to
another device or a WR or NOCOP to any device is issued, the retire will take place. Figure 18
(right) illustrates a situation in which the controller wants to issue a WR-WR-RD COLC packet
sequence, with all commands addressed to the same device, but addressed to any combination of
banks and columns.
The RD will prev ent a retire of the f irst WR from automatically happening. But the first dualoct D (a1)
in the write buffer w ill be ov erwr itten by the second W R dualoct D(b1) if the RD command is issued
in the third COLC packet. Therefore, it is required in this situation that the controller issue a NOCOP
command in the third COLC packet, delay ing the RD command by a time of tPACKET. This situation
is explicitly s hown in Table 13 for the cases in which tCCDELAY is equal to tRTR.
Figure 18 Retire Held Off b y Read (left) and Controller Forces WWR Gap (right)
Figure 19 shows a possible result when a retire is held of f for a long t ime (an extend ed version of
Figure 18-lef t). After a WR command, a series of six RD commands ar e issued to the same device
(but to any combination of bank and column addresses). In the meantime, the bank Ba to which the
WR command was originally directed is precharged, and a different row Rc is activated. When the
retire is automatically performed, it is m ade to this new row, si nce the w rite buffer only contains t he
bank and column address, not the row address. The controller can insure that this doesn’t happen
by never precharging a bank with an unretired write buffer. Note that in a system with more than one
RDRAM, there will nev er be more than two RDRAMs with unretired write buffers. This is because
DQB8...0
DQA8...0
t
Transaction a: WR
Transaction b: RD
DQA8...0
DQB8...0
a1 = {Da, Ba, Ca1}
b1 = {Da, Bb, Cb1}
PACKETRTR
CWD
t
+
t
D (a1) Q
c1 = {Da, Bc, Cc1}
b1 = {Da, Bb, Cb1}
a1 = {Da, Ba, Ca1}
t
CWD
Transaction c: RD
Transaction b: WR
Transaction a: WR
t
D (a1)
RTR
D (b1)
SPA04222
ROW2...
CTM/CFM
COL4...
ROW0
COL0
T5
The retire operation for a write can be
held off by a read to the same device
CTM/CFM
COL0
ROW2...
ROW0
COL4...
WR a1
T0 T1 T2 T4T3
CAC
retire (a1)
RD b1 MSK (a1)
t
T15T10T6 T7 T9T8 T11 T12 T14T13 T20T16 T17 T19T18
The controller must insert a NOCOP to retire (a1)
to make room for the data (b1) in the write buffer
WR a1 WR b1 retire (a1)
MSK (a1) RD c1
T0 T1 T3T2 T4 T5 T6 T9T8T7 T10 T11 T14T13T12 T15 T16
CAC
t
T19T18T17 T20
Data Sheet 35 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
a WR command issued to one device automatically retires the write buffers of all other devices
written a time tRTR before or earlier.
Figure 19 Retire Held Off b y Reads to Same D evice, Write Buffer Retired to New Row
CWD
t
CAC
t
a0 = {Da, Ba, Ra}
b4 = {Da, Bb, Cb4}
b1 = {Da, Bb, Cb1}
DQA8...0
DQB8...0
Transaction a: WR
Transaction c: WR
Transaction b: RD
c0 = {Da, Ba, Rc} b5 = {Da, Bb, Cb5}
b2 = {Da, Bb, Cb2}
a1 = {Da, Ba, Ca1}
D (a1)
This sequence is hazardous
Q (b2)
b6 = {Da, Bb, Cb6}
b3 = {Da, Bb, Cb3}
Q (b1)
a2 = {Da, Ba}
with caution
and must be used
WARNING
Q (b3) Q (b4)
SPT04223
Q (b5)
T24
ACT a0
ROW2...
ROW0
COL4...COL0
CTM/CFM
T2T0 T1 T3 T4
t
WR a1 RD b1 RD b2
RAS
t
RC
T14T7T5 T6 T8 T9 T10 T11 T13T12 T19T15 T16 T18T17 T20 T21 T23T22
MSK (a1)
retire (a1)
write data in the new row
The retire operation puts the
ACT c0
RD b5
RTR
t
RCD
t
RD b3 RD b4
PRER a2
RP
t
RD b6
T34T29T25 T26 T27 T28 T30 T31 T32 T33 T39T36T35 T37 T38 T41T40 T42 T43 T46T45T44 T47
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 36 6.00
Interleaved Write - Example
Figure 20 shows an example of an interleaved write transaction. Transactions similar to the one
presented in Figure 16 are directed to non-adjacent banks of a single RDRAM. This allows a new
transaction to be issued once every tRR interval rather than once every tRC interval (four tim es more
often). The DQ data pin efficiency is 100% with this sequence.
With two dualoc ts of data written per t ransaction, the COL, DQA, and D QB pins are fully utilized.
Banks are precharged using the WRA autoprecharge option rat her than the PRER comm and in an
ROWR packet on t he ROW pins.
In this example, the first transaction is directed to device Da and bank Ba. The next three
transactions are directed to the same dev ice Da, but need t o us e dif ferent, non-adjacent banks Bb,
Bc, Bd so there is no bank conflict. The fifth transaction could be redirected back to bank Ba without
interference, since the first transaction would have completed by then (tRC has elapsed). Each
transaction may use any value of row address (Ra, Rb, …) and column address (Ca1, Ca2, Cb1,
Cb2, …).
Figure 20 Interleaved Write Transaction with Two Dualoct Data Length
y1 = {Da, Ba+4, Cy1}
z1 = {Da, Ba+6, Cz1}
a1 = {Da, Ba, Ca1}
b1 = {Da, Ba+2, Cb1}
c1 = {Da, Ba+4, Cc1}
d1 = {Da, Ba+6, Cd1}
e1 = {Da, Ba, Ce1}
f1 = {Da, Ba+2, Cf1}
Transaction y: WR
Transaction z: WR
Transaction a: WR
Transaction b: WR
Transaction c: WR
Transaction d: WR
Transaction e: WR
Transaction f: WR
y0 = {Da, Ba+4, Ry}
z0 = {Da, Ba+6, Rz}
a0 = {Da, Ba, Ra}
b0 = {Da, Ba+2, Rb}
c0 = {Da, Ba+4, Rc}
d0 = {Da, Ba+6, Rd}
e0 = {Da, Ba, Re}
f0 = {Da, Ba+2, Rf}
SPA04224
f3 = {Da, Ba+2}
e3 = {Da, Ba}
d3 = {Da, Ba+6}
c3 = {Da, Ba+4}
b3 = {Da, Ba+2}
z3 = {Da, Ba+6}
y3 = {Da, Ba+4}
a3 = {Da, Ba}
f2 = {Da, Ba+2, Cf2}
e2 = {Da, Ba, Ce2}
d2 = {Da, Ba+6, Cd2}
c2 = {Da, Ba+4, Cc2}
b2 = {Da, Ba+2, Cb2}
z2 = {Da, Ba+6, Cz2}
y2 = {Da, Ba+4, Cy2}
a2 = {Da, Ba, Ca2}
T24
D (a1)
ACT a0
ROW0
ROW2...
COL4...COL0
DQB8...0
DQA8...0
D (y1)
WR z1
CTM/CFM
T2T1T0 T3 T4
MSK (y2)
D (y2)
RCD
t
WRA z2 WR a1
ACT b0
D (z2)D (z1)
t
CWD
ACT c0
T7T6T5 T8 T9 T12T11T10 T13 T14 T17T15 T16 T18 T19
RC
t
T22T20 T21 T23 T44
D (c2)
Transaction e can use the
same bank as transaction a
D (a2) D (b1)
ACT d0
t
RR
D (c1)D (b2)
ACT e0 ACT f0
T34T29T27T25 T26 T28 T32T30 T31 T33 T39T37T35 T36 T38 T40 T41 T43T42 T45 T46 T47
D (x2) D (d1)
MSK (y1) MSK (z1) WR a2
MSK (z2) WR b1
MSK (a1) WR b2
MSK (a2) WR c1
MSK (b1) WR c2
MSK (b2) WR d1
MSK (c1) WR d2
MSK (c2) WR e1
MSK (d1) WR e2
MSK (d2)
Data Sheet 37 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Interleaved Read - Example
Figure 21 shows an example of interleaved read transactions. Transactions similar to the one
presented in Figure 15 are directed to non-adjacent banks of a single RDRAM. The address
sequence is identical to the one used in t he previous w rite example. The DQ data pins efficiency is
also 100%. The only difference with the write example (aside from the use of the RD command
rather than the WR command) is the use of the PREX command in a COLX packet to precharge the
banks rather than the RDA command. This is done because the PREX is available for a read
transaction but is not available for a masked write transaction.
Interleaved RRWW - Example
Figure 22 shows a steady-state sequence of 2-dualoct RD/RD/WR/WR… transactions directed to
non-adjacent banks of a single RDRAM. This is similar to the interleaved write and read examples
in Figure 20 and Figure 21 except that bubble cycles need to be inserted by the controller at
read/write boundaries. The DQ data pin efficiency for the example in Figure 22 is 32/42 or 76%. If
there were more RDRAMs on the Channel, t he DQ pin ef ficiency w ould approach 32/34 or 94% for
the two-dualoct RRWW sequence (this case is not shown).
In Figure 22, the first bubble type tCBUB1 is inserted by the controller between a RD and WR
command on the COL pins . This bubble accounts for the round-trip propagation delay that is seen
by read data, and is explained in detail in Figure 4. This bubble appears on the DQA and DQB pins
as tDBUB1 between a w rite data dualoct D and read data dualoct Q. This bubble also appears on the
ROW pins as tRBUB1.
Figure 21 Interleaved Read Transaction with Two Dualoct Data Length
y1 = {Da, Ba+4, Cy1}
z1 = {Da, Ba+6, Cz1}
a1 = {Da, Ba, Ca1}
b1 = {Da, Ba+2, Cb1}
c1 = {Da, Ba+4, Cc1}
d1 = {Da, Ba+6, Cd1}
e1 = {Da, Ba, Ce1}
f1 = {Da, Ba+2, Cf1}
Transaction y: RD
Transaction z: RD
Transaction a: RD
Transaction b: RD
Transaction c: RD
Transaction d: RD
Transaction e: RD
Transaction f: RD
y0 = {Da, Ba+4, Ry}
z0 = {Da, Ba+6, Rz}
a0 = {Da, Ba, Ra}
b0 = {Da, Ba+2, Rb}
c0 = {Da, Ba+4, Rc}
d0 = {Da, Ba+6, Rd}
e0 = {Da, Ba, Re}
f0 = {Da, Ba+2, Rf}
SPT04225
f3 = {Da, Ba+2}
e3 = {Da, Ba}
d3 = {Da, Ba+6}
c3 = {Da, Ba+4}
b3 = {Da, Ba+2}
z3 = {Da, Ba+6}
y3 = {Da, Ba+4}
a3 = {Da, Ba}
f2 = {Da, Ba+2, Cf2}
e2 = {Da, Ba, Ce2}
d2 = {Da, Ba+6, Cd2}
c2 = {Da, Ba+4, Cc2}
b2 = {Da, Ba+2, Cb2}
z2 = {Da, Ba+6, Cz2}
y2 = {Da, Ba+4, Cy2}
a2 = {Da, Ba, Ca2}
T24
RD b2
Q (a1)
PREX a3
ACT a0
ROW0
ROW2...
COL4...COL0
DQB8...0
DQA8...0
Q (x2)
RD z1
CTM/CFM
T2T1T0 T3 T4
PREX z3
PREX y3
Q (y1) Q (y2)
RCD
t
RD z2 RD a1
ACT b0
Q (z2)Q (z1)
t
CAC
ACT c0
RD b1
RD a2
T7T6T5 T8 T9 T12T11T10 T13 T14 T17T15 T16 T18 T19
RC
t
T22T20 T21 T23 T44
RD e1
Q (c2)
Transaction e can use the
same bank as transaction a
PREX b3
Q (a2) Q (b1)
RD c1
ACT d0
t
RD c2
RR
PREX c3
Q (c1)Q (b2)
RD d2RD d1
ACT e0 ACT f0
T34T29T27T25 T26 T28 T32T30 T31 T33 T39T37T35 T36 T38 T40 T41 T43T42
PREX
Q (d1)
RD e2
T45 T46 T47
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 38 6.00
The second bubble type tCBUB2 is inserted (as a NOCOP command) by the controller between a WR
and RD command on the COL pins when there is a WR-WR-RD sequence to the same device. This
bubble enables write data to be retired f rom the write buf fer without being los t, and is explained in
detail in Figure 18. There would be no bubble if address c0 and address d0 were directed to
different devices. This bubble appears on the D QA and DQB pins as tDBUB2 between a write data
dualoct D and read data dualoct Q. This bubble also appears on t he ROW pins as tRBUB2.
Figure 22 Interleaved RRWW Sequ ence with Two Dualoct Data Length
y1 = {Da, Ba+4, Cy1}
z1 = {Da, Ba+6, Cz1}
a1 = {Da, Ba, Ca1}
b1 = {Da, Ba+2, Cb1}
c1 = {Da, Ba+4, Cc1}
d1 = {Da, Ba+6, Cd1}
e1 = {Da, Ba, Ce1}
f1 = {Da, Ba+2, Cf1}
Transaction z: RD
Transaction a: RD
Transaction d: RD
Transaction e: RD
Transaction y: WR
Transaction b: WR
Transaction c: WR
Transaction f: WR
y0 = {Da, Ba+4, Ry}
z0 = {Da, Ba+6, Rz}
a0 = {Da, Ba, Ra}
b0 = {Da, Ba+2, Rb}
c0 = {Da, Ba+4, Rc}
d0 = {Da, Ba+6, Rd}
e0 = {Da, Ba, Re}
f0 = {Da, Ba+2, Rf}
SPT04226
f3 = {Da, Ba+2}
e3 = {Da, Ba}
d3 = {Da, Ba+6}
c3 = {Da, Ba+4}
b3 = {Da, Ba+2}
z3 = {Da, Ba+6}
y3 = {Da, Ba+4}
a3 = {Da, Ba}
e2 = {Da, Ba, Ce2}
a2 = {Da, Ba, Ca2}
f2 = {Da, Ba+2, Cf2}
d2 = {Da, Ba+6, Cd2}
c2 = {Da, Ba+4, Cc2}
b2 = {Da, Ba+2, Cb2}
z2 = {Da, Ba+6, Cz2}
y2 = {Da, Ba+4, Cy2}
WRA b2
PREX a3
T5
ACT a0
ROW0
DQB8...0
DQA8...0
COL4...COL0
t
D (y2)
DBUB1
RD z1
CBUB2
t
ROW2...
CTM/CFM
T0 T1 T2 T3 T4
ACT b0 ACT c0
RD a2
PREX z3
Q (z1)
DBUB2
t
RD a1
RD z2
Q (z2) Q (a1)
CBUB1
WR b1
t
RBUB2
t
T10T6 T7 T8 T9 T12T11 T13 T14 T17T16T15 T18 T19 T22T21T20 T23 T24
Transaction e can use the
same bank as transaction a
RD d0
ACT e0ACT d0
D (b2)D (b1)Q (a2)
WR c1
MSK (b1) MSK (b2)
WRA c2
D (c1) D (c2)
DBUB1
t
NOCOP
MSK (c2)
CBUB2
MSK (c1)
NOCOP
t
RBUB2
t
T27T26T25 T28 T29 T32T31T30 T33 T34 T37T36T35 T38 T39 T42T40 T41 T43 T44 T47T45 T46
MSK (y2)
Data Sheet 39 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Control Register Transactions
The RDRAM has two CMOS input pins SCK and CMD and t wo CMOS input/output pins SIO0 and
SIO1. These provide serial access to a set of control registers in the RDRAM. These control
registers provide configuration information to the controller during the initialization process. They
also allow an application to select the appropriate operating mode of the RDRAM.
SCK (serial clock) and CMD (command) are driven by the controller to all RDRAMs in parallel. SIO0
and SIO1 are connected (in a daisy chain fashion) from one RDRAM to the next. In normal
operation, the data on SIO0 is repeated on SIO1, which connects to SIO0 of the next RDRAM (the
data is repeated from SIO1 to SIO0 for a read data packet). The controller connects to SIO0 of t he
first RD RA M.
Figure 23 Serial Write (SWR) Transaction to Control Register
Write and read transactions are each composed of four packets, as shown in Figure 23 and
Figure 24. Each packet consists of 16 bits, as summarized in Figure 15 and Figure 16. The packet
bits are sampled on the falling edge of SCK. A transaction begins with a SRQ (Serial Request)
packet. This packet is framed w ith a 11110000 pattern on the CMD input (note that the CMD bits are
sampled on both the falling edge and the rising edge of SCK). The SRQ packet contains the
SOP3…SOP0 (Serial Opcode) field, which selects the transaction type. The SDEV5…SDEV0
(Serial Device address) selects one of the 32 RDRAMs. If SBC (Serial Broadcast) is set, then all
RDRAMs are selected. The SA (Serial Addres s) packet contains a 12 bit address for selecting a
control register.
A write transaction has a SD (Serial Data) packet next. This contains 16 bits of data that is written
into the selected control regis ter. A SINT (Serial Interval) packet is last, providing some delay for
any side-effects to take place. A read transaction has a SINT packet, then a SD packet. This
provides delay for the selected RDRAM to access the control register. The SD read data packet
travels in the opposite direc tion (towards the controller) from the ot her packet types. The SCK cycle
time will accommodate the total delay.
SPT04227
T36
SIO1
1111
SIO0
CMD
0000
SCK
T4
Each packet is repeated
SRQ - SWR Command
from SIO0 to SIO1
00000000...00000000
SRQ-SWR Command
SA
00000000...00000000
SA
T20
SD
00000000...00000000
SD
SINT
SINT
00000000...00000000
T52
1
1
0
0
1
1111
0
0
T68
1
Next transaction
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 40 6.00
Figure 24 Serial Read (SRD) Transaction Control Register
SRQ - SRD Command
SIO1
First 3 packets are repeated
from SIO0 to SIO1
SA
Non-addressed RDRAMs pass
0/SD15...SD0/0 from SIO1 to SIO0
SINT
0
SD
SPT04228
1
0
0
0
Addressed RDRAM drives
0/SD15...SD0/0 on SIO0
00000000...00000000
SRQ-SRD Command
1111
SIO0
CMD
0000
SCK
T4 T20
00000000...00000000
SA
T36
Next transaction
T52
00000000...00000000
SINT
0
SD
0 on SIO0
00000000...00000000
Controller drives
1
0
1
0
0
T68
1
1111
Data Sheet 41 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Control Register Packets
Figure 25 SETR, CLRR, SETF Transaction
Table 15 summarizes the formats of the four packet types for control register transactions.
Table 16 sum marizes the fields that are used within the packets.
Figure 25 shows the transaction format for the SETR, CLRR, and SETF commands. These
transactions consist of a single SRQ packet, rather than four packets like the SWR and SRD
commands. The same framing sequence on the CMD input is used, however. These commands are
used during init ialization prior to any control register read or write transactions.
Table 15 C ontrol Register Packet Formats
SCK
Cycle SIO0 or
SIO1
for
SRQ
SIO0 or
SIO1
for SA
SIO0 or
SIO1
for
SINT
SIO0 or
SIO1
for SD
SCK
Cycle SIO0 or
SIO1
for
SRQ
SIO0 or
SIO1
for SA
SIO0 or
SIO1
for
SINT
SIO0 or
SIO1
for SD
0 rsrv rsrv 0 SD15 8 SOP1 SA7 0 SD7
1 rsrv rsrv 0 SD14 9 SOP0 SA6 0 SD6
2 rsrv rsrv 0 SD13 10 SBC SA5 0 SD5
3 rsrv rsrv 0 SD12 11 SDEV4 SA4 0 SD4
4 rsrv SA11 0 SD11 12 SDEV3 SA3 0 SD3
5 SDEV5 SA10 0 SD10 13 SDEV2 SA2 0 SD2
6 SOP3 SA9 0 SD9 14 SDEV1 SA1 0 SD1
7 SOP2 SA8 0 SD8 15 SDEV0 SA0 0 SD0
SPT04229
The packet is repeated
from SIO0 to SIO1
SRQ packet - SETR/CLRR/SETF
SIO1
1
0
0
SRQ packet-SETR/CLRR/SETF
1111
SIO0
CMD
0000
SCK
T4
00000000...00000000 1
1
0
0
T20
1
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 42 6.00
Table 16 F ield Description for Control Register Packets
Field Description
rsrv Reserved. Should be driven as “0” by c ontroller.
SOP3 … SOP0 0000 - SRD. Serial read of control register {SA11 … SA0} of RDRAM
{SDEV5 … SDEV0}.
0001 - SWR. Serial write of c ontrol register {SA11 … SA0} of RD R AM
{SDEV5 … SDEV0}.
0010 - SETR . Set Reset bit, all control registers assume their reset values.1)
16 tSCYCLE delay until CLRR command.
0100 - SETF . Set fast (normal) clock mode. 4 tSCYCLE delay until next
command.
1011 - CLRR. Clear Reset bit, all control registers retain their reset values.1)
4tSCYCLE delay until next command.
1111 - NOP. No serial operation.
0011, 0101-1010, 1100-1110 - RSRV. Reserved encodings.
SDEV5 …
SDEV0 Serial device. Compared to SDEVID5…SDEVID0 field of INIT control register
field to selec t the RDRAM to which the transaction is directed.
SBC Serial broadcast. When set, RDRAMs ignore {SDEV5 … SD EV0} for
RDRAM selection.
SA11…SA0 Serial address. Selects which control r egister of the selected RD RAM is read
or written.
SD15…SD0 Serial data. The 16 bits of data written to or read from the selected control
register of the selected RDRAM.
1) The SETR and CLRR comma nds must always be applied in two succe ssive tr ansactio ns to RDRAMs; i .e. they
may not be used in isolation. This is called “SETR/CLRR Reset”.
Data Sheet 43 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Initialization
Figure 26 SIO Reset Sequen ce
Initialization refers to the process that a controller must go through after power is applied to the
system or the system is reset. The controller prepares the RDRAM s ub-system for norm al C hannel
operation by (primarily) using a sequence of control register transactions on the serial CMOS pins.
The following steps outline the sequence seen by the various memory subsystem components
(including the RDRAM components) during initialization. This sequence is available in the form of
reference code. Contact Rambus Inc. for more information.
1. Start Clocks – This step calculates the proper clock frequencies for PClk (controller logic),
SynClk (RAC block), RefClk (DRCG component), CTM (RDRAM component), and SCK (SIO
block).
2. R AC In itiali zati on – This step causes the INIT block to generate a sequence of pulses which
resets the RAC, performs RAC maintenance operations, and measures tim ing intervals in order
to ensure clock stability.
3. RDRAM In itializati on – This stage performs most of the steps needed to initialize the RDRAMs.
The rest are performed in stages 5.0, 6.0, and 7.0. All of the steps in 3.0 are carried out through
the SIO block interface.
3.1./3.2. SIO Reset – T his reset operation is performed before any SIO c ontrol register read or
write transactions. It clears six registers (TEST34, CCA, CCB, SKIP, TEST78, and
TEST79) and places the INIT register into a special s tate (all bits cleared except SKP and
SDEVID fields are set to ones).
3.3. Write TEST77 Register – The TEST77 register must be explicitly written with zeros before
any other regist ers are read or written.
3.4. Write TCYCLE Register – The TC YCLE register is written with the cycle time tCYCLE of the
CTM clock (for Channel and RDRAMs) in units of 64ps. The tCYCLE v alue is determined in
stage 1.0.
3.5. Write SDEVID Register – The SDEVID (serial device identification) register of each
RDRAM is written with a unique address value so that directed SIO read and write
transactions can be performed. This address value inc reases from 0 to 31 according to the
distance an RDRAM is from the ASIC component on the SIO bus (the closest RDRAM is
address 0).
SPT04230
from SIO0 to SIO1
The packet is repeated
SIO1
0000000000000000
1
0
0
00001100
SIO0
CMD
SCK
T0
00000000...00000000
0000000000000000
1
1
0
0
T16
1
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 44 6.00
3.6. Write DEVID Register – The DEVID (device identification) register of each RDRAM is
written with a unique address value so that directed memory read and write transactions
can be performed. This address value increases from 0 to 31. The DEVID value is not
necessarily the same as the SDEVID v alue. RDRAMs are sorted into regions of the same
core configuration (number of bank, row, and column address bits and core type).
3.7. Write PDNX,PDNXA Registers – The PD NX and PDNXA registers are writ ten with values
that are used to measure the timing intervals connected with an exit from the PDN
(powerdown) power state.
3.8. Write NAPX Reg ister – The NAPX register is written with values that are used to measure
the timing intervals connected with an exit from the NAP power state.
3.9. Write TPARM R egister – The TPARM register is w ritten with values which determine the
time interval betw een a COL packet with a memory read command and the Q pac ket with
the read data on the Channel. T he values written set each RDRAM to the minimum value
permitted for the system. This will be adjusted later in s tage 6.0.
3.10. Write TCDLY1 Register – The TCDLY1 register is written with values which determine the
time interval betw een a COL packet with a memory read command and the Q pac ket with
the read data on the Channel. T he values written set each RDRAM to the minimum value
permitted for the system. This will be adjusted later in s tage 6.0.
3.11. Write TFRM Register – The T FRM register is w ritten with a value that is related to the tRCD
parameter for the sy stem. The tRCD parameter is the time interval between a ROW packet
with an act ivate command and the COL packet with a read or write command.
3.12. 3.12 SETR/CLRR – Each RDRAM is given a SETR command and a CLRR command
through the SIO block. T his sequence performs a second reset operation on the RDRAMs.
3.13. Write CCA and CCB Registers - These registers are written with a value halfway between
their minimum and maximum values. T his s hortens the time needed for the RDRAMs to
reach their st eady-state current control values in s tage 5.0.
3.14. Powerdown Exit – The RDRAMs are in the PDN power state at this point. A broadcast
PDNExit command is performed by the SIO block to place the RDRAMs in the RLX (relax)
power state in which they are ready to receive ROW packets.
3.15. SETF - Each RDRAM is given a SETF command through the SIO block. One of the
operations performed by this step is to generate a value for the AS (autoskip) bit in the SKIP
register and fix the RDRAM to a particular read domain.
4. Controller Confi gurati on – This stage initializes the controller block. Each step of this stage will
set a field of the ConfigRMC[63:0] bus to the appropriate value. Other controller implementations
will have similar initialization requirements, and this stage may be used as a guide.
4.1. Initial Read Data Offset – The ConfigRMC bus is written w ith a value which determines
the time interval bet ween a COL packet with a memory read command and the Q packet
with the read data on the Channel. The value written sets RMC.d1 to the minimum value
permitted for the system. This will be adjusted later in s tage 6.0.
4.2. Configure Row/Column Timing – This step determines the values of the tRAS,MIN, tRP,MIN,
tRC,MIN, tRCD,MIN, tRR,MIN, and tPP,MIN RDRAM timing parameters that are present in the
system. The ConfigRMC bus is written with values that will be compatible with all RDRAM
devices that are present.
4.3. Set Refresh Interval – This step determines the values of the tREF,MAX RDRAM timing
parameter that are present in the system. The ConfigRMC bus is written with a value that
will be c ompatible with all RDRAM devices that are present.
Data Sheet 45 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
4.4. Set Current Control Interval – This step determines the values of the tCCTRL,MAX RDRAM
timing parameter that are present in the system. The ConfigRMC bus is written with a value
that will be compatible with all R DRAM devices that are present.
4.5. Set Slew Rate Control Interval – This step determines the values of the tTEMP,MAX RDRAM
timing parameter that are present in the system. The ConfigRMC bus is written with a value
that will be compatible with all R DRAM devices that are present.
4.6. Set Bank/Row/Col Address Bits – This step determines the number of RDRAM bank,
row, and column address bits that are present in the system. It also determines the RDRAM
core types (independent , doubled, or split) t hat are pres ent. T he C onfigRMC bus is w ritten
with a value that will be c ompatible with all RDRAM devices that are present.
5. RDRAM Current Control – T his step causes the INIT block to generate a sequence of pulses
which performs RDRAM maintenance operations.
6. RDRAM Core, Read Domain Initialization – This stage completes the RDRAM initialization
6.1. RDRAM Core Initialization – A sequence of 192 memory refresh transactions is
performed in order to place the cores of all RDRAMs into the proper operating state.
6.2. RDRAM Read Domain Initialization - A memory write and memory read transaction is
performed to each RD RAM to determine which read domain each RDRAM occupies. The
programmed delay of each RDRAM is then adjusted so the total RDRAM read delay
(propagation delay plus programmed delay) is constant. The TPARM and TCDLY1
registers of each RDRAM are rewritten with the appropriate read delay values. The
ConfigRMC bus is also rewritten with an updated value.
7. Other RDRAM Register Fields – This stage rewrites the INIT register with the final values of the
LSR, NSR, and PSR fields.
In essence, the controller mus t read all the read-only configuration registers of all RDRAMs (or it
must read the SPD device present on each RIMM), it must process this information, and then it must
write all the read-write registers to place the RDRAMs into the proper operating m ode.
Initialization Note [1]:
During the initialization process, it is necessary for the controller to perform 128 current control
operations (3xCAL, 1xCAL/SAM) and one temperature calibrate operation (TCEN/TCAL) after reset
or after powerdown (PDN) exit.
Initialization Note [2]:
There are two classes of 288Mbit RDRAM. They are distinguished by the “S28IECO” bit in the SPD.
The behavior of the RDRAM at initializ ation is slightly different for the tw o types:
S28IECO = 0: Upon powerup the device enters ATTN state. The serial operations SETR, CLRR,
and SETF are performed without requiring a SDEVID match of the SBC bit
(broadcast) to be set .
S28IECO = 1: Upon powerup the device enters PDN state. The serial operations SETR, CLRR,
and SETF require a SDEVID match.
The INFINEON 288Mbit RDRAM supports the sec ond versio with S28IECO = 1.
See the document detailing the reference initialization procedure for more information on how to
handle this in a system.
Initialization Note [3]:
After the step of equalizing the t otal read delay of each RDRAM has been c ompleted (i.e. after the
TCDLY0 and TCDLY1 fields have been written for the final time), a single final memory read
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 46 6.00
transaction should be made to each RDRAM in order to ensure that the output pipeline stages have
been cleared.
Initialization Note [4]:
The SETF command (in t he serial SRQ packet) s hould only be issued onc e during the Initialization
process, as should the SETR and CLR R commands.
Initialization Note [5]:
The CLRR command (in the serial SRQ pac ket) leaves some of the contents of the memory core in
an indeterminate state.
Control Register Summary
Table 17 s ummarizes the RDRAM control registers. Detail is provided for each control register in
Figure 27 through Figure 43. Read-only bit s which are shaded gray are unused and return z ero.
Read-write bits which are shaded gray are reserved and should always be written with zero. The
RIMM SPD Application Note (DL-0054) describes additional read-only configuration regis ters which
are present on Direct RIMMs.
The state of the register fields are potentially affected by the IO Reset operat ion or the SETR/CLRR
operation. This is indicated in the text accompanying each register diagram.
Table 17 C ontrol Register Summary
SA11…SA0 Register Field read-write/
read-only Description
02116 INIT SDEVID read-write, 6 bits Serial device ID. Device address for control register
read/write.
PSX read-write, 1 bit Power select exit. PDN/NAP exit w ith device addr on
DQA5 … 0.
SRP read-write, 1 bit SIO repeater. Used to initialize RDRAM.
NSR read-write, 1 bit NAP self-refresh. Enables self-refresh in NAP mode.
PSR read-write, 1 bit PDN self-refresh. Enables self-refresh in PDN mode.
LSR read-write, 1 bit Low power self-refresh. Enables low power
self-refresh.
TEN read-write, 1 bit Temperature sensing enable.
TSQ read-write, 1 bit Temperature sensing output.
DIS read-write, 1 bit RDRAM disable.
02216 TEST34 TEST34 read-write, 16 bits Test register. Do not read or write after SIO reset.
02316 CNFGA REFBIT read-only, 3 bit Refresh bank bits. Used for multi-bank refresh.
DBL read-only, 1 bit Double. Specifies doubled-bank architecture
MVER read-only, 6 bit Manufacturer version. Manufacturer identification
number.
PVER read-only, 6 bit Protocol version. Specifies version of Direct protocol
supported.
Data Sheet 47 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
02416 CNFGB BYT read-only, 1 bit Byte. Specifies an 8-bit or 9-bit byte size.
DEVTYP read-only, 3 bit Device type. Device can be RDRAM or some other
device category.
SPT read-only, 1 bit Split-core. Each core half is an individual dependent
core.
CORG read-only, 6 bit Core organization. Bank, row, column address field
sizes.
SVER read-only, 6 bit Stepping version. Mask version number.
04016 DEVID DEVID read-write, 5 bits Device ID. Device address for memory read/write.
04116 REFB REFB read-write, 4 bits Refresh bank. Next bank to be refreshed by
self-refresh.
04216 REFR REFR read-write, 9 bits Refresh row. Next row to be refreshed by REFA, self-
refresh.
04316 CCA CCA read-write, 7 bits Current control A. Controls IOL output current for DQA.
ASYMA read-write, 2 bits Asymmetry control. Controls asymmetry of VOL/VOH
swing for DQA.
04416 CCB CCB read-write, 7 bits Current control B. Controls IOL o u t put current for DQB.
ASYMB read-write, 2 bits Asymmetry control. Controls asymmetry of VOL/VOH
swing for DQB.
04516 NAPX NAPXA read-write, 5 bits NAP exit. Specifies length of NAP exit phase A.
NAPX read-write, 5 bits NAP exit. Specifies length of NAP exit phase
A + phase B.
DQS read-write , 1 bits DQ select. Selects CMD framing for N AP/PDN exit.
04616 PDNXA PDNXA read-write, 13 bits PDN exit. Specifies length of PDN exit phase A.
04716 PDNX PDNX read-write, 13 bits PDN exit. Specifies length of PDN exit phase
A + phase B.
04816 TPA RM TCAS read-write, 2 bits tCAS-C core parameter. Dete rmines tOFFP data sheet
parameter.
TCLS read-write, 2 bits tCLS-C core parameter. Determines tCAC and tOFFP
parameters.
TCDLY0 read-write, 3 bits tCDLY0-C c or e parame ter . Prog rammabl e delay for r ead
data.
04916 TFRM TFRM read-write, 4 bits tFRM-C core parameter. Determines ROW-COL packet
framing interval.
04a16 TCDLY1 TC DLY1 read-write, 3 bits tCDLY1-C cor e parame t er. Prog rammabl e delay for r ead
data.
04c16 TCYCLE TCYCLE read-write, 14 bits tCYCLE data sheet par ameter. Specifies cycle time in
64 ps units.
Table 17 C ontrol Register Summary (cont’d)
SA11…SA0 Register Field read-write/
read-only Description
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 48 6.00
04b16 SKIP AS read-only, 1 bit Autoskip value established by the SETF command.
MSE read-write, 1 bit Manual skip enable. Allows the MS value to override
the AS value.
MS read-write, 1 bit Manual skip value.
04d16- TEST77 TEST77 read-write, 16 bits Test register. Write with zero after SIO reset.
04e16- TEST78 TEST78 read-write, 16 bits Test register. Do not read or write after SIO reset.
04f16- TEST79 TEST79 read-write, 16 bits Test register. Do not read or write after SIO reset.
08016 - 0ff16 reserved reserved vendor-specific Vendor-specific test registers. Do not read or write
after SIO reset.
Table 17 C ontrol Register Summary (cont’d)
SA11…SA0 Register Field read-write/
read-only Description
Data Sheet 49 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 27 INIT Register
0
15 14131211109876543210
Control Register: INIT
SDE
VID
5
DIS TSQ TEN LSR PSR NSR SRP PSX 0 SDEVID4...SDEVID0
SPD04273
Address: 021
16
PSX - Power Exit Select. PDN and NAP are exited with (=0) or without (=1)
a device address on the DQA5..0 pins. PDEV5 (on DQA5) selectes broadcast (1)
or directed (0) exit. For a directed exit, PDEV4..0 (on DQA4..0) is compared to
DEVID4..0 to select a device.
S
RP - SIO Repeater. Controls value on SIO; SIO1 = SIO0 if SRP = 1,
SIO1 = 1 if SRP = 0 SRP resets to 1.
NAP Self-Refresh. NSR = 1 enables self-refresh in NAP mode.
NSR can’t be set while in NAP mode. NSR resets to 0.
PDN Self-Refresh. PSR=1 enables self-refresh in PDN mode.
PSR can’t be set while in PDN mode. PSR resets to 0.
Low Power Self-Refresh. LSR = 1 enables longer self-refresh interval.
The self-refresh supply current is reduced. LSR resets to 0.
Temperature Sensing Enable. TEN = 1 enables temperature sensing circuitry,
permitting the TSQ bit to be read to determine if a thermal trip point has been
exceeded. TEN resets to 0.
Temperature Sensing Output. TSQ = 1 when a temperature trip point has been
exceeded, TSQ = 0 when it has not. TSQ is available during a current control
operation (see Figure 51).
RDRAM Disable. DIS = 1 causes RDRAM to ignore NAP/PDN exit sequence,
DIS = 0 permits normal operation. This mechanism disables an RDRAM.
DIS resets to 0.
SDEVID5..0 - Serial Device Identification. Compared to
SDEV5..0 serial address field of serial request packet for
register read/write transactions. This determines which
RDRAM is selected for the register read or write operation.
SDEVID resets to 3F
16
.
Read/write register.
Reset values are undefined except as affected by SIO
Reset as noted below. SETR/CLRR Reset does not
affect this register.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 50 6.00
Figure 28 CNFGA Register
Figure 29 CNFGB Register
15 14 13 12 11 10 9876543210
Control Re g i ster: CNF GA Address: 02316
0000000000000000
Read-only register.
REFBIT2..0 - Refresh Bank Bit s. Speci fies the number of bank
add ress bits used by RE FA and REF P com m a nd s. Perm i ts
mul ti-bank refresh in f uture RDRAMs.
DBL - Doubled-Bank. DBL=1 means the device uses a doubled-
bank arch itecture with adjacent-bank d ependency. DBL=0
mea ns no dependency.
MVER5..0 - Manufacturer Version. Specifies the manufacture r
identification number.
PVER5..0 - Protocol Version. Specifies the Direct Protocol
version used by this device:
0 - Compliant with version 0.62
1 - Compliant with version 0.7
2 - Version 1 plus Interleaved Device Mode
3 to 63 Re served.
The INFIN E ON 288 M Bit RD R AM suppo rts Version 1 plus
Interleaved Device Mode
REFBIT2..0
= 100
PVER5..0
= 000001 DBL
1
MVER5..0
= mmmmmm
Note: In RDRAMs with protocol version 1 PVER[5:0] = 000001,
the range of the PDNX field (PDNX[2:0] in the PDNX register)
may not be large enough to specify the location of the restricted
interval in Figure 47. In this case, the effective tS4 parameter must
increase and no row or column packets may overlap the restricted
interval. See Figure 47 and Table 18.
15 14131211109876543210
Control Register: CNFGB
SVER5...0
= ssssss CORG4...0
= xxxxx DEVTYP2...0
= 000
SPT
1BYT
B
Address: 024
16
DEVTYP2..0 - Device type. DEVTYP = 000 means
that this device is an RDRAM.
BYT - Byte width. B = 1 means the device reads
and writes 9-bit memory bytes. B = 0 means 8 bits.
CORG4..0 - Core organization. This field specifies
the number of bank (3, 4, 5, or 6 bits), row
(9, 10, 11, or 12 bits), and column (5, 6, or 7 bits)
address bits. The encoding of this field will be
specified in a later version of this document.
SVER5..0 - Stepping version. Specifies the mask
version number of this device.
SPT - Split-core. SPT = 1 means the core is split,
SPT = 0 means it is not.
Read-only register.
SPD04255
Data Sheet 51 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 30 TEST Register
Figure 31 DEVID Register
11
0
14
00
15 0 0
13 12
Address: 022
16
Control Register: TEST34
0 0
10 9 00 0
786 3
00 0
54 0 0
210
0
Read/write registers.
Reset value of TEST34 is zero ( from SIO Reset).
This register are used for testing purposes. It must
not be read or written after SIO Reset.
SPD04276
11
0
14
00
15 0 0
13 12
Address: 040
16
Control Register: DEVID
0 0
10 9 00 0
786 3
0
54
DEVID4...DEVID0
210
Read/write register.
Reset value is undefined.
Device Identification register.
DEVID4..DEVID0 is compared to DR4..DR0,
DC4..DC0, and DX4..DX0 fields for all memory
read or write transactions. This determines which
RDRAM is selected for the memory read or write
transaction.
SPD04277
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 52 6.00
Figure 32 REFB Register
Figure 33 CCA Register
11
0
14
00
15 0 0
13 12
Address: 041
16
Control Register: REFB
0 0
10 9 00 0
786 3
00
5420
Read/write register.
Reset value is zero (from SETR/CLRR).
Refresh Bank register.
REFB4..REFB0 is the bank that will be refreshed
next during self-refresh. REFB4..0 is incremented
after each self-refresh activate and precharge
operation pair.
REFB4...REFB0
1
SPD04256
11
0
14
00
15 0 0
13 12
Address: 043
16
Control Register: CCA
0 0
10 9
ASYMA
0
0786 35420
Read/write register.
Reset value is zero (SETR/CLRR or SIO Reset).
CCA6...CCA0 - Current Control A. Controls the
I
OL
output current for the DQA8..DQA0 pins.
ASYMB0 control the asymmetry of the
V
OL
/
V
OH
voltage swing about the
V
REF
reference voltage
for the DQA8...0 pins.
CCA6...CCA01
SPD04279
Data Sheet 53 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 34 REFR Register
Figure 35 CCB Register
11
0
14
00
15 0 0
13 12
Address: 042
16
Control Register: REFR
0 0
10 9 786 35420
Read/write register.
Reset value is zero (from SETR/CLRR).
Refresh Row register.
REFR8...REFR0 is the row that will be refreshed
next by the REFA command or by self-refresh.
REFR8...0 is incremented when BR4...0 = 1111 for
the REFA command. REFR8...0 is incremented
when REFB4...0 = 1111 for self-refresh.
REFR8...REFR0 1
SPD04257
Read/write register.
Reset value is zero (SETR/CLRR or SIO Reset).
CCB6...CCB0 - Current Control B. Controls the
I
OL
output current for the DQB8...DQB0 pins.
ASYMB0 control the asymmetry of the
V
OL
/
V
OH
voltage swing about the
V
REF
reference voltage
for the DQB8...0 pins.
11
0
14
00
15 0 0
13 12
Address: 044
16
Control Register: CCB
0 0
10 9
ASYMB
0
0786 35420
CCB6...CCB01
SPD04281
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 54 6.00
Figure 36 NAPX Register
Figure 37 PDNXA Register
15 14131211109876543210
Control Register: NAPX
NAPX4..0 NAPXA4..0
Address: 045
16
NAPXA4...0 - Nap Exit Phase A. This field speci-
fies the number of SCK cycles during the first
phase for exiting NAP mode. It must satisfy:
NAPXA *
tSCYCLE
≥
t
NAPXA,MAX
Do not set this field to zero.
NAPX4...0 - Nap Exit Phase A plus B. This field specifies
the number of SCK cycles during the first plus second
phases for exiting NAP mode. It must satisfy:
NAPX *
tSCYCLE
≥ NAPXA *
tSCYCLE
+
t
NAPXB,MAX
Do not set this field to zero.
DQS - DQ Select. This field specifies the number of SCK
cycles (0 => 0.5 cycles, 1 => 1.5 cycles) between the
CMD pin framing sequence and the device selection
on DQ5...0. See Figure 48 - This field must be written
with a “1” for this RDRAM.
Read/write register.
Reset value is undefined
Note -
tSCYCLE
is
tCYCLE1
(SCK cycle time).
0
DQS
0000
SPD04282
1114
00
15 0
13 12
Address: 046
16
Control Register: PDNXA
10 9 PDNXA12...0
786 354 210
Read/write register.
Reset value is undefined
PDNXA4...0 - PDN Exit Phase A. This field specifies
the number of (64 * SCK cycle) units during
the first phase for exiting PDN mode. It must
satisfy:
PDNXA * 64 *
t
SCYCLE ≥
t
PDNXA, MAX
Do not set this field to zero.
Note - only PDNXA5...0 are implemented.
Note -
t
SCYCLE is
t
CYCLE1 (SCK cycle time).
SPD04283
Data Sheet 55 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 38 PDNX Register
Figure 39 TPARM Register
1114
00
15 0
13 12
Address: 047
16
Control Register: PDNX
10 9 PDNX12...0
786 354 210
Read/write register.
Reset value is undefined
PDNX4...0 - PDN Exit Phase A plus B. This field
specifies the number of (256 * SCK cycle) units
during the first plus second phases for exiting
PDN mode. It should satisfy:
PDNX * 256 *
t
SCYCLE ≥ PDNXA * 64 *
t
SCYCLE +
t
PDNXB, MAX
If this equation can’t be satisfied, then the
maximum PDNX value should be written, and
the
t
S4
/
t
H4
timing window will be modified (seeFigure 49).
Do not set this field to zero.
Note - only PDNX2...0 are implemented.
Note -
t
SCYCLE is
t
CYCLE1 (SCK cycle time).
SPD04284
11
0
14
00
15 0 0
13 12
Address: 048
16
Control Register: TPARM
0 0
10 9 00 786 3
TCDLY0
54TCLS
21
TCAS
0
Read/write register.
Reset value is undefined.
TCAS1..0 - Specifies the
t
CAS-C core parameter in
t
CYCLE units. This should be “10” (2 *
t
CYCLE ).
TCLS1..0 - Specifies the
tCLS-C
core parameter in
t
CYCLE units. Should be “10” (2 *
t
CYCLE).
TCDLY0 - Specifies the
t
CDLY0-C core parameter in
t
CYCLE units. This adds a programmable delay to
Q (read data) packets, permitting round trip read
delay to all devices to be equalized. This field may
be written with the values “010” (2 *
t
CYCLE )
through “101” (5 *
t
CYCLE ).
The equations relating the core parameters to the
datasheet parameters follow:
t
CAS-C = 2 *
t
CYCLE
t
CLS-C = 2 *
t
CYCLE
t
CPS-C = 1 *
t
CYCLE Not programmable
t
OFFP =
t
CPS-C +
t
CAS-C +
t
CLS-C - 1 *
t
CYCLE
= 4 *
t
CYCLE
t
RCD =
t
RCD-C + 1 *
t
CYCLE -
t
CLS-C
=
t
RCD-C - 1 *
t
CYCLE
t
CAC = 3 *
t
CYCLE +
t
CLS-C +
t
CDLY0-C +
t
CDLY1-C
(see table below for programming ranges)
TCDLY0 TCDLY1
010 000 not allowed
t
CDLY0-C
t
CDLY1-C 7 *
t
CYCLE
0 *
t
CYCLE
2 *
t
CYCLE
011 000 8 *
t
CYCLE
0 *
t
CYCLE
3 *
t
CYCLE
011 001 9 *
t
CYCLE
1 *
t
CYCLE
3 *
t
CYCLE
011 010 10 *
t
CYCLE
2 *
t
CYCLE
3 *
t
CYCLE
100 010 11 *
t
CYCLE
2 *
t
CYCLE
4 *
t
CYCLE
101 010 12*
t
CYCLE
8 *
t
CYCLE
9 *
t
CYCLE
10 *
t
CYCLE
11 *
t
CYCLE
12*
t
CYCLE
2 *
t
CYCLE
5 *
t
CYCLE
t
CAS
@
t
CYCLE = 3.3 ns
t
CAS
@
t
CYCLE = 2,5 ns
SPD04285
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 56 6.00
Figure 40 TFRM Register
Figure 41 TRDLY Register
11
0
14
00
15 0 0
13 12
Address: 049
16
Control Register: TFRM
0 0
10 9 00 0
786 3
0 0
5420
Read/write register.
Reset value is undefined.
TFRM3...0 - Specifies the position of the framing
point in
t
CYCLE units. This value must be greater
or equal to the
t
FRM,MIN
parameter. This is
the minimum offset between a ROW packet
(which places a device at ATTN) and the first
COL packet (directed to that device) which must
be framed. This field may be written with the
values “0111” (7 *
t
CYCLE ) through “1010”
(10 *
t
CYCLE ). TFRM is usually set to the value
which matches the largest
tRCD,MIN
parameter
(modulo 4 *
t
CYCLE
) that is present in an RDRAM
in the memory system. Thus, if an RDRAM with
t
RCD, MIN = 11 *
t
CYCLE were present, then TFRM
would be programmed to 7 *
t
CYCLE .
TFRM3...0
1
SPD04286
11
0
14
00
15 0 0
13 12
Address: 04A
16
Control Register: TCDLY1
0 0
10 9 00 0
786 3
000
5420
Read/write register.
Reset value is undefined.
TCDLY1 - Specifies the value of the
t
CDLY1-C core
parameter in
t
CYCLE units. This adds a programm-
able delay to Q (read data) packets, permitting
round trip read delay to all devices to be equalized.
This field may be written with the values
“000” (0 *
t
CYCLE ) through “010” (2 *
t
CYCLE ). Refer
to Figure 39 for more details.
TCDLY1
1
SPD04287
Data Sheet 57 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 42 SKIP Register
Figure 43 TEST Registers
11
MSE
14
00
15 0
AS
13 12
Address: 04B
16
Control Register: SKIP
MS
0
10 9 00 0
786 3
00 0
54 0 0
210
0
Read/write register (except AS field).
Reset value is zero (SIO Reset).
AS - Autoskip. Read-only value determined by
autoskip circuit and stored when SETF serial
command is received by RDRAM during initial-ization.
In figure 58, AS = 1 corresponds to the
early Q(a1) packet and AS = 0 to the Q(a1) packet
one
t
CYCLE later for the four uncertain cases.
MSE - Manual skip enable (0 = auto, 1 = manual).
MS - Manual skip (MS must be 1 when MSE = 1).>
During initialization, the RDRAMs at the furthest
point in the fifth read domain may have selected
the AS = 0 value, placing them at the closest point
in a sixth read domain. Setting the MSE/MS fields
to 1/1 overrides the autoskip value and returns
them to the furthest point of the fifth read
domain.
SPD04288
11
0
14
00
15 0 0
13 12
Address: 04F
16
Control Register: TEST79 Address: 04E
16
Control Register: TEST78 Address: 04D
16
Control Register: TEST77
0 0
10 9 00 0
786 3
00 0
54 0 0
210
0
Read/write registers.
Reset value of TEST78, 79 is zero ( SIO Reset).
Do not read or write TEST78, 79 after SIO reset.
TEST77 must be written with zero after SIO reset.
These registers must only be used for testing
purposes.
SPD04289
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 58 6.00
Figure 44 TCYCLE Register
Power State Management
Table 18 sum marizes the power states available to a Direct RDRAM. In general, the lowest power
states have the longest operational latencies. For example, the relative power levels of PDN state
and STBY state have a ratio of about 1:110, and the relative access latencies to get read data have
a ratio of about 250:1.
PDN state is the lowest power state available. The information in the RDRAM core is usually
maintained with self-refresh; an internal timer automatically refreshes all rows of al l banks. PDN has
a relatively long exit latency because the TCLK/RCLK block must resynchronize itself to the external
clock signal.
NAP state is another low-power state in which either self-refresh or REFA-refresh are used to
maintain the core. See “Refresh” on page 64 for a description of the two refresh mechanisms. NAP
has a shorter exit lat ency than PDN because the TCLK/RCLK block maintains its synchronization
state relative to the external clock signal at the time of NAP entry. This imposes a limit (tNLIMIT) on
how long an RDRAM may remain in NAP state before briefly returning to STBY or ATTN to update
this synchronization state.
Table 18 Po wer State Summary
Power
State Description B locks Consuming
Power Power
State Description Blocks Consuming
Power
PDN Powerdown state. Self-refresh NAP Nap state. Similar to
PDN except lower
wake-up latency.
Self-refresh or
REFA-refresh
TCLK/RCLK-Nap
STBY Standby state.
Ready for ROW
packets.
REFA-refresh
TCLK/RCLK
ROW demux receiver
ATTN Attention state.
Ready for ROW and
COL packets.
REFA-refresh
TCLK/RCLK
ROW demux receiver
COL demux receiver
ATTNR Attention read state.
Ready for ROW and
COL packets.
Sending Q (read
data) packets.
REFA-refresh
TCLK/RCLK
ROW demux receiver
COL demux receiver
DQ mux transmitter
Core power
ATTNW Attention write state.
Ready for ROW and
COL packets.
Ready for D (write
data) packets.
REFA-refresh
TCLK/RCLK
ROW demux receiver
COL demux receiver
DQ demux receiver
Core power
1114
00
15 13 12
Address: 04C
16
Control Register: TCYCLE
10 9
TCYCLE13...TCYCLE0
786 354 210
Read/write register.
Reset value is undefined
TCYCLE13...0 - Specifies the value of the
t
CYCLE
datasheet parameter in 64 ps units. For the
t
CYCLE, MIN of 2.5 ns (2500 ps), this field should be
written with the value “00027 16 ” (39 * 64 ps).
SPD04290
Data Sheet 59 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 45 summarizes the transition conditions needed for moving between the various power
states. Note that NAP and PDN have been divided into two substates (NAP-A/NAP-S and
PDN-A/PDN-S) to account for the fact that a NAP or PDN exit may be made to either ATTN or STBY
states.
At initialization, the SETR/CLRR Reset sequence will put the RDRAM into PDN-S state. The PDN
exit sequence involves an optional PDEV spec ification and bits on the CMD and SIOIN pins.
Once the RDRAM is in STBY, it will move to the ATTN/ATTNR/ATTNW states when it receives a
non-broadcast ROWA packet or non-broadcast ROWR packet with the ATTN command. The
RDRAM returns to STBY f rom these three states when it receives a RLX command. Alternatively,
it may enter NAP or PDN state from ATTN or STBY states with a NAPR or PDNR command in an
ROWR packet. The PDN or NAP exit sequence involves an optional PDEV specification and bits on
the CMD and SIO0 pins. The RDRAM returns to the ATTN or STBY state it was originally in when
it first entered NAP or PDN.
An RDRAM may only rem ain in NAP state for a time tNLIMIT. It must periodically return to AT TN or
STBY.
The NAPRC command causes a napdown operation if the RDRAM’s NCBIT is set. T he NCBIT is
not directly visible. It is undefined on reset. It is set by a N APR command t o t he RDRAM, and it is
cleared by an AC T command to the R DRAM. I t permits a controller t o m anage a s et of RDRAMs in
a mixture of power states.
STBY state is the normal idle state of the RDRAM. In this s tat e all banks and sense amps have
usually been left precharged and ROWA and ROWR packets on the ROW pins are being
monitored. When a non-broadcast ROWA packet or non-broadcast ROWR packet (with the ATTN
command) packet addressed to the RDRAM is seen, the RDRAM enters ATTN state (see the right
side of Figure 46). This requires a tim e tSA during which the RDRAM activates the specified row of
the specified bank. A time TFRM × tCYCLE after the ROW packet, the RDRAM will be able to frame
COL packets (TFRM is a control regist er field - see Figure 40). Once in ATTN s tate, t he R DRAM
will automatically transition to the ATTNW and ATTNR states as it receives WR and RD commands.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 60 6.00
Figure 45 Power State Transition Diagram
Once the RDRAM is in ATTN, ATTNW, or ATTNR states, it will remain there until it is explicitly
returned to the ST BY state with a RLX command. A RLX command may be given in an ROWR,
COLC, or COLX packet (see the left side of Figure 46). It is usually given after all banks of the
RDRAM have been precharged; if other banks are still activated, then the RLX command would
probably not be given.
If a broadcast ROWA packet or ROWR packet (with the ATTN command) is received, the RDRAM’s
power state doesn’t change. If a broadcast ROWR packet with RLXR command is received, the
RDRAM goes to STBY.
SPD04231
ATTNR
ATTN
Automatic
Automatic
Automatic
Automatic ATTNW
NAP-A
Automatic
Automatic
NAP-S
NAP
PDEV.CMD x SIO0
NAPR x RLXR
NAPR x RLXR
PDEV.CMD x SIO0
STBY
PDN-S
PDN-A
PDN
PDNR
NAPR
ATTN
PDNR x RLXR
PDEV.CMD x SIO0
PDNR x RLXR
PDEV.CMD x SIO0
RLX
SETR/CLRR
NLIMIT
t
Notation:
SETR/CLRR
PDNR
NAPR
RLXR
RLX
SIO0
PDEV.CMD
ATTN (Non-Broadcast) with ATTN Command
- SETR/CLRR Reset Sequence in SRQ Packets
- ROWA Packet (Non-Broadcast) or ROWR Packet
- RLX Command in ROWR, COLC, COLX Packets
- RLX Command in ROWR Packet
- NAPR Command in ROWR Packet
- PDNR Command in ROWR Packet
- (PDEV = DEVID) x (CMD = 01)
- SIO0 Input Value
Data Sheet 61 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 47 shows the NAP entry sequence (left). NAP state is entered by sending a NAPR
command in a ROW packet. A time tASN is required to ent er NAP state (t his specification is provided
for power c alculation purposes). The clock on C TM/CFM m ust rem ain s table for a t ime tCD after the
NAPR command.
The RDRAM may be in ATTN or STBY state when the NAPR command is issued. When NAP state
is exited, the RDRAM will return to t he original starting state (ATTN or ST BY). If it is in ATTN state
and a RLXR c ommand is specified with NAPR, then the RDRAM w ill return to STBY state when
NAP is exit ed.
Figure 47 also shows the PDN entry sequence (right). PDN st ate is entered by sending a PDNR
command in a ROW packet. A t ime tASP is required to enter PDN s tate (this specification is prov ided
for power c alculation purposes). The clock on C TM/CFM m ust rem ain s table for a t ime tCD after the
PDNR command.
The RDRAM may be in ATTN or STBY s tate when the PDNR command is issued. When PDN s tate
is exited, the RDRAM will return to t he original starting state (ATTN or ST BY). If it is in ATTN state
and a RLXR command is s pecified with PDNR, then the RDRAM will return to STBY state when
PDN is exited. The current- and slew -rate-control levels are re-established.
The RDRAM’s write buffer must be retired with t he appropriate COP command before NAP or PDN
are entered. Also, all the RDRAM’s banks must be precharged before NAP or PDN ar e entered. The
exception to this is if NAP is entered with the NSR bit of the INIT register cleared (disabling
self-refresh in NAP). The commands for relaxing, retiring, and precharging may be given to the
RDRAM as late as the ROPa0, C OPa0, and XOPa0 packets in Figure 47. No broadcast packets
nor packets directed to the RDRAM entering Nap or PDN may overlay the quiet window. This
window extends for a time tNPQ after the packet with the NAPR or PDNR command.
Figure 48 shows the NAP and PDN exit sequences. These sequences are virtually identical; the
minor differences will be highlighted in the following description.
Before NAP or PDN exit, the CTM/CFM clock m ust be stable f or a time tCE. Then, on a falling and
rising edge of SCK, if there is a “01” on the CMD input, NAP or PDN state will be exited. Also, on the
falling SCK edge the SIO0 input mus t be at a 0 for N AP exit and 1 for PDN exit.
If the PSX bit of the INIT register is 0, then a device PDEV5 … 0 is specified for NAP or PDN exit
on the DQA5 … 0 pins. This value is driven on the rising SCK edge 0.5 or 1.5 SCK cycles after the
original falling edge, depending upon the value of the DQS bit of t he NAPX register. If t he PSX bit
of the INIT regis ter is 1, then the RDRAM ignores the PDEV5 … 0 address packet and ex its NAP or
PDN when the wake-up sequence is presented on the CMD wire. The ROW and COL pins must be
quiet at a time tS4/tH4 around the indicated falling SCK edge (timed with the PDNX or NAPX register
fields). After that, ROW and COL packets may be directed to the RDRAM which is now in ATTN or
STBY state.
Figure 49 shows the constraints for entering and exiting NAP and PDN states. On the left side, an
RDRAM exits NAP st ate at the end of cycle T3. This RDRAM may not re-enter NAP or PDN state
for an interval of tNU0. The RDRAM ent ers NAP state at the end of cycle T13. This RDRAM may not
re-exit NAP state for an interval of tNU1. The equations for these two parameters depend upon a
number of factors, and are shown at the bottom of the f igure. NAPX is the value in the NAPX field
in the NAPX register.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 62 6.00
Figure 46 STBY Entry (left) and STBY Exit (right)
Figure 47 NAP Entry (left) and PDN Entry (right)
On the right side of Figure 48, an RDRAM exits PDN state at the end of cycle T3. This RDRAM may
not re-enter PD N or N AP s tate f or an interval of tPU0. The RDRAM enters PDN state at the end of
cycle T13. This RDRAM may not re-exit PDN state for an interval of tPU1. The equations for these two
parameters depend upon a num ber of factors, and are shown at the bottom of the figure. PDNX is
the value in the PDNX field in the PDNX register.
at (TFRM)
ATTN
Power
State
t
AS
STBY
State
DQB8...0
DQA8...0
Power
t
CYCLE
SPT04232
device (d1!= d0) is okay
A COL packet to another
STBY ATTN
SA
t
or earlier.
at (TFRM - 4) *
t
*
CYCLE
ROP a0
CTM/CFM
ROW2...
ROW0
COL4...
COL0
DQB8...0
DQA8...0
RLXX
RLXC
RLXR
T0 T3T1 T2 T4 T5
COL4...
COL0
ROW2...
CTM/CFM
ROW0
T8T6 T7 T9 T10 T13T11 T12 T14 T0 T1 T2
A COL packet to device d0
(or any other device) is okay
a1 = {d1, b1, c1}
ROWA or ROWR/ATTN
ROP = non-broadcast
No COL packets may be
placed in the three
indicated positions; i.e. at
(TFRM - {1, 2, 3})
COP a0
XOP a0
TFRM
COP a1
XOP a1
*t
CYCLE
a0 = {d0, b0, r0}
T5T4T3 T6 T7 T10T9T8 T11 T12 T15T14T13 T16
CYCLE
t
*
or later.
owerlap the restricted
The (eventual) NAP / PDN exit will be to the same ATTN / STBY state the RDRAM was in prior to NAP / PDN entry
Power
State
a)
DQA8...0
DQB8...0
NAP
a)
ATTN / STBY
ASN
t
State
Power
DQA8...0
DQB8...0
SPT04233
the restricted interval will
directed to device d0 after
ROW or COL packets
ATTN / STBY
a)
ASP
t
PDN
interval
be ignored
ROP a0
COP a0
XOP a0
(NAPR)
COL0
COL4...
ROW0
ROW2...
CTM/CFM
T0 T1 T2 T3
Restricted ROP a1
Restricted
NPQ
t
XOP a1
COP a1
t
CD
COL0
COL4...
ROW0
CTM/CFM
ROW2...
T6T5T4 T7 T8 T11T10T9 T12 T13 T14 T0
No ROW or COL packets
directed to device d0 may
overlap the restricted
interval. No broadcast
ROW packets may overlap
the quiet interval
device other than d0 may
ROW or COL packets to a
a1 = {d1, b1,r1, c1}
a0 = {d0, b0, r0, c0}
COP a1
XOP a1
ROP a1
Restricted
(PDNR)
Restricted
XOP a0
COP a0
NPQ
t
ROP a0
CD
t
T3T1 T2 T4 T5 T8T6 T7 T9 T10 T13T11 T12
Data Sheet 63 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 48 NAP and PD N Exit
t
Device selection timing slot is selected by DQS field of NAPX register
asserted at NAP or PDN entry time
The DQS field must be written with "1" for this RDRAM
Exit to STBY or ATTN depends upon whether RLXR was
d)
c)
DQS = 0
from SIO0 to SIO1
NAP / PDN
The packet is repeated
Use 0 for NAP exit, 1 for PDN exit
State
Power
SIO1
b)
a)
0/1
SIO0
CMD
SCK
0/1
0
CE
DQS = 1
bb
a
a
1
DQS = 0
b
SPT04234
d
STBY / ATTN
*
+ (PDNXA
is exiting the NAP-A or PDN-A states
restricted interval if device PDEV
No COL packets may overlap the
Effective hold becomes
T24T23T21T20T19T18T16T15 T17T14T13T10 T11 T12T9T8T5 T6 T7T4T3T2T1 T22T0
t
If PSX = 1 in Init register, then NAP/PDN
exit is broadcast (no PDEV field).
COL4...COL0
DQA8...0
DQB8...0
ROW0
CTM/CFM
ROW2...
DQS = 1
b, c
S3
H3
t
S3
tt
H3
< (PDNXA
if (PDNX
tt
=
H4H4
*64
256
*
T47T46T45T42 T43T41T40T38T36T35 T37 T39T31T30T28T26T25 T27 T29 T34 T44
XOP
COP
ROP
COP Restricted
PDNXB, MAX
PDNXB, MAX
SCYCLE
t
*
64
*
SCYCLE
SCYCLE
*
t
t
+
t
)
t
+
S4
)
t
XOP
H4
t
the restricted interval
No ROW packets may overlap
S4
t
H4
t
(NAPX
SCYCLE
t
*)
*
SCYCLE
t
) / (256 PDNX*
- (PDNX *256 *
SCYCLE
t
)
ROP Restricted
).
´
PDEV5...0PDEV5...0
bb
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 64 6.00
Figure 49 NAP Entry/Exit Windows (left) and PDN Entry/Exit Windows (right)
Refresh
RDRAMs, like any ot her DRAM technology, use volatile storage cells which must be periodically
refreshed. This is accomplished with the REF A command. Figure 50 shows an example of this.
The REFA command in the transaction is typically a broadcast command (DR4T and DR4F are both
set in the ROWR packet), so that in all devices bank number Ba is activated with row number REFR,
where REFR is a control regis ter in the RDRAM. When the command is broadcast and AT TN is set,
the power state of the RDRAMs (ATTN or STBY) will remain unchanged. The controller increments
the bank address Ba for the next REFA command. When Ba is equal to its maximum value, the
RDRAM automatically inc rements REFR for the next REFA command.
On average, these REFA commands are sent once every tREF/2BBIT+RBIT (where BBIT are the
number of bank address bits and RBIT are the number of row address bits) so that each row of each
bank is refres hed once every tREF interval.
The REFA command is eq uivalent to an ACT command, in terms of the way that it interacts with
other packets (see Table 12). In the example, an ACT command is sent after tRR to address b0, a
different (non-adjacent) bank than the REFA c ommand.
A second ACT command can be sent after a time tRC to address c0, the same bank (or an adjacent
bank) as the REFA command.
Note that a broadcast REFP command is issued a time tRAS after the initial REFA command in order
to precharge the refreshed bank in all R DRAMs. After a bank is given a REF A command, no other
core operations (activate or precharge) should be issued to it until it receives a REFP.
It is also possible to interleave refresh transactions (not shown). In the figure, the ACT b0 command
would be replaced by a REFA b0 command. The b0 address would be broadcast to all devices, and
PDN exitNAP exit
+ (2 + NAPX)
no entry to NAP or PDN
if NSR = 1
CMD
5
8
23
NU1
NU1
t
t
=
=
NU0
t
=
CYCLE
CYCLE
CYCLE
*
*
tt
*
t
- (0.5
01
SCYCLE
) if NSR = 0
SCYCLE
t
*
t
*
NU0
t
NU1
no exit
t
0
CMD
SCYCLE
PDNX)
no entry to NAP or PDN
SCYCLE
SCYCLE
SCYCLE
5
8
23
PU1
PU1
t
t
=
=
PU0
t
=
t
*
*
t
*
t
0
*
if PSR = 1
+ (2 + 256
- (0.5
t
*
1
PU0
t
SPT04235
PU1
no exit
) if PSR = 0
*
SCYCLE
t
t
0
ROW2...
CTM/CFM
ROW0
ROW2...
CTM/CFM
SCK
T2T0 T1 T3 T4 T6T5 T7 T18T13
NAP entry
T10T8 T9 T11 T12
NAPR
T14 T15 T17T16 T19
SCK
ROW0
T10T2 T3 T4 T6T5 T7 T9T8
PDN entry
NAPR
T13T11 T12 T14 T15 T17T16 T18 T19
Data Sheet 65 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
would be {Broadcast,Ba+2,REFR}. Note that the bank address should skip by two to avoid adjacent
bank interference. A possible bank inc rementing pattern w ould be: {13, 11, 9, 7, 5, 3, 1, 8, 10, 12,
14, 0, 2, 4, 6, 15, 29, 27, 25, 23, 21, 19, 17, 24, 26, 28, 30, 16, 18, 20, 22, 31}. Every time bank 31
is reached, t he REFA command would automatically increment the REFR register.
A second refresh mechanism is available for us e in PD N and NAP power states. T his mechanism
is called self-refresh mode. When the PDN power state is entered, or when NAP power state is
entered with the NSR control register bit set, then self-refresh is automatically started for the
RDRAM.
Self-refresh uses an internal time base reference in the RDRAM. This causes an activate and
precharge to be carried out once in every tREF/2BBIT+RBIT interval. The REFB and REFR control
registers are used to keep track of the bank and row being refreshed.
Before a controller places an RDRAM into self-refresh mode, it should perform REFA/REFP
refreshes until the bank address is equal to the maximum value. This ensures that no rows are
skipped. Likewise, when a c ontroller returns an RDRAM to REFA/REFP refresh, it should start with
the minimum bank address value (zero).
Figure 50 REFA/REFP Refresh Tr ansaction Example
Current and T emperature Control
Figure 51 shows an example of a transaction which performs current control calibration. It is
necessary to perform this operation once to every RDRAM in ev ery tCCTRL interval in order to keep
the IOL output current in its proper range.
BBIT + RBIT
/2
a1 = {Broadcast, Ba}
DQA8...0
DQB8...0
Transaction a: REFA
Transaction a: REFA
Transaction c: XX
Transaction b: XX
d0 = {Broadcast, Ba+1, REFR}
c0 = {Dc, ==Ba, Rc}
b0 = {Db, /={Ba, Ba+1, Ba-1}, Rb}
a0 = {Broadcast, Ba, REFR}
t
REF
REFR = REFR8...REFR0
REFB = REFB3...REFB0
BBIT = #row address bits
BBIT = #bank address bits
SPT04236
T24
REFP a1
CTM/CFM
COL4...COL0
ROW2...
ROW0
REFA a0
T2T0 T1 T3 T4
t
RR
ACT b0
t
RAS
t
RC
T14T7T5 T6 T8 T9 T12T10 T11 T13 T19T17T15 T16 T18 T20 T21 T23T22 T44
REFA d0
ACT c0
RP
t
T34T29T25 T26 T28T27 T30 T31 T33T32 T35 T36 T37 T41 T42 T43 T46T45 T47
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 66 6.00
This example uses f our COLX packets with a CAL command. These cause the RDRAM to drive four
calibration packets Q(a0) a t ime tCAC later. An of fset of tRDTOCC must be placed bet ween the Q(a0)
packet and read data Q(a1)from the same device. These calibration packets are driven on the
DQA4 … 3 and DQB4 … 3 wires. The TSQ bit of the INIT register is driven on the DQA5 wire during
same interval as the calibration packets. The remaining DQA and DQB wires are not used during
these calibration packets. The las t COLX packet also contains a SAM command (concatenated with
the CAL command). The RDRAM samples the last calibration packet and adjusts its IOL current
value.
Unlike REF commands, CAL and SAM commands cannot be broadcast. This is because the
calibration packets from dif ferent devices would int erfere. Therefore, a current control transaction
must be sent every tCCTRL/N, where N is the number of RDRAMs on the Channel. The device field
Da of the address a0 in the CAL/SAM command should be incremented after each transaction.
Figure 23 shows an ex ample of a t emperature calibration sequence to t he RDRAM. This sequence
is broadcast once every tTEMP interval to all the RDRAMs on the Channel. The TCEN and TCAL are
ROP commands, and cause the slew rate of the output drivers to adjust for t emperature drift. During
the quiet interv al tTCQUIET the devices being c alibrated can’t be read, but they can be written.
Figure 51 Current Control CAL/SAM Transaction Example
HotTemp = DQA5
Note that DQB3 could be used instead of DQA3.
When used for monitoring, it should be enabled with the DQA3
bit (current control one value) in case there is no RDRAM present:
control register; i.e. logic 0 or high voltage means hot temperature.
DQA5 of the first calibrate packet has the inverted TSQ bit of INIT
READTOCC
Transaction a2: CAL/SAM
Transaction a0: CAL/SAM
Transaction a1: RD
COL4...COL0
DQB8...0
DQA8...0
Q (a1)
t
CAL a0
a1 = {Da, Bx}
a2 = {Da, Bx}
CAL/SAM a0
Q (a0)
a0 = {Da, Bx}
CAL a0
t
CAC
CAL a0 CAL a2
DQA3
*
Q (a1)
CCSAMTOREAD
t
SPT04237
T24
packet position or earlier.
command must be at this
device from an earlier RD
Read data from the same
ROW0
ROW2...
CTM/CFM
T2T0 T1 T3 T4 T14
device from an earlier RD
prior to the Q(a0) packet.
Read data from a different
command can be anywhere
T5 T6 T8T7 T9 T10 T11 T13T12
t
T19T15 T16 T18T17 T20 T21 T23T22
packet position or later.
device from a later RD
command must be at this
Read data from a different
T34
Read data from a different
command can be anywhere
after to the Q(a0) packet.
CCTRL
T29
device from a later RD
T25 T26 T27 T28 T31T30 T32 T33 T39T36T35 T37 T38 T42 T43 T46T45T44 T47
Data Sheet 67 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 52 Temperature Calibration (TCEN-TCAL) Transactions to RDRAM
t
being calibrated
No read data from devices
DQB8...0
DQA8...0
TCAL
SPT04238
T24
TCQUIET
Any ROW packet may be plased in the gap between the
ROW packets with the TCEN and TCAL commands.
CTM/CFM
ROW2...
COL4...COL0
ROW0
TCEN
T2T0 T1 T3 T4
TCAL
t
TCEN
t
T14T5 T6 T8T7 T9 T10 T11 T13T12 T19T15 T16 T18T17 T20 T21 T23T22
TCEN
TEMP
t
T34T25 T26 T27 T28 T32 T33 T36T35 T37 T38 T41 T42 T43 T46T45T44 T47
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 68 6.00
Table 19 El ectrical Conditions
Parameter and Co nditions Symbol Limit Values Unit
min. max.
Junction temperature under bias TJ- 100 °C
Supply voltage VDD, VDDA 2.50 – 0.13 2.50 + 0.13 V
Supply voltage droop ( DC) during N AP int erval
(tNLIMIT)VDD,N, VDDA,N –2.0%
Supply voltage ripple (AC) during NAP interval
(tNLIMIT)VDD,N, VDDA,N –2.0 2.0 %
Supply voltage f or CMOS pins (2.5 V
controllers)
Supply voltage f or CMOS pins (1.8 V
controllers)
VCMOS 2.50 – 0.13
1.80 – 0.1 2.50 + 0.25
1.80 + 0. 2 V
V
Termination voltage VTERM 1.80 – 0.1 1.80 + 0.1 V
Reference voltage VREF 1.40 – 0.2 1.40 + 0.2 V
RSL data input - low voltage VDIL VREF – 0.5 VREF – 0.2 V
RSL data input - high voltage VDIH VREF + 0.2 VREF + 0.5 V
RSL data input swing: VDIS = VDIH – VDIL VDIS 0.4 1.0 V
RSL data asy mmetry: ADI = [(VDIH – VREF) +
(VDIL – VREF)]/VDIS
ADI 0 – 20 %
RSL clock input - crossing point of true and
complement signals VX1.3 1.8 V
RSL clock input - common mode
VCM =(VCIH +VCIL)/2 VCM 1.4 1.7 V
RSL clock input swing: VCIS =VCIH –VCIL
(CTM, CTMN pins). VCIS,CTM 0.35 0.70 V
RSL clock input swing: VCIS =VCIH –VCIL
(CFM, CFMN pins). VCIS,CFM 0.225 0.70 V
CMOS input low voltage VIL,CMOS – 0.3 VCMOS/
2 – 0.25 V
CMOS input high voltage VIH,CMOS VCMOS/
2 + 0.25 VCMOS
+0.3 V
Data Sheet 69 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Ta ble 20 Tim ing Cond itions
Parameter Symbol Limit Values Unit F igure
min. max.
CTM and CFM cycle times (-600)
CTM and CFM cycle times (-711)
CTM and CFM cycle times (-800)
tCYCLE 3.33
2.80
2.50
3.83
3.83
3.83
ns
ns
ns
Figure 53
Figure 53
Figure 53
CTM and CFM input rise and f all times tCR, tCF 0.2 0.5 ns Figure 53
CTM and CFM high and low tim es tCH, tCL 40% 60% tCYCLE Figure 53
CTM-CFM differential (MSE/MS = 0/0)
CTM-CFM differential (MSE/MS = 1/1)1) tTR 0.0
0.9 1.0
1.0 tCYCLE Figure 42
Figure 53
Domain crossing window tDCW – 0.1 0.1 tCYCLE Figure 59
DQA/DQB/ROW/COL input rise/fall times tDR, tDF 0.2 0.65 ns Figure 54
DQA/DQB/ROW/COL-to-CFM set/hold @
tCYCLE =3.33ns
DQA/DQB/ROW/COL-to-CFM set/hold @
tCYCLE =2.81ns
DQA/DQB/ROW/COL-to-CFM set/hold @
tCYCLE =2.50ns
tS, tH0.2752),d
0.2403),4)
0.200d
–
–
–
ns
ns
ns
Figure 54
Figure 54
Figure 54
SIO0, SIO1 input rise and fall t imes tDR1, tDF1 –5.0nsFigure 56
CMD, SCK input rise and fall t imes tDR2, tDF2 –2.0nsFigure 56
SCK cycle tim e - Serial control register
transactions tCYCLE1 1000 – ns Figure 56
SCK cycle tim e - Power transitions 10 – ns Figure 56
SCK high and low times tCH1, tCL1 4.25 – ns Figure 56
CMD setup t ime to SCK rising or f alling
edge5) tS1 1.00 – ns Figure 56
CMD hold t ime to SCK rising or f alling
edgectH1 1–nsFigure 56
SIO0 setup t ime to SCK falling edge tS2 40–nsFigure 56
SIO0 hold t ime to SCK falling edge tH2 40–nsFigure 56
PDEV setup time on DQA5 … 0 to SCK
rising edge. tS3 0–nsFigure 48,
Figure 57
PDEV hold time on DQA5 … 0 to SCK
rising edge. tH3 5.5 – ns
ROW2 … 0, COL4 … 0 setup time for
quiet window tS4 – 1 – tCYCLE Figure 48
ROW2 … 0, COL4 … 0 hold time for quiet
window6) tH4 5–tCYCLE Figure 48
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 70 6.00
CMOS input low voltage -
over/undershoot voltage duration is less
than or equal to 5 ns
VIL,CMOS – 0.7 VCMOS/
2 – 0.4 V–
CMOS input high voltage -
over/undershoot voltage duration is less
than or equal to 5 ns
VIH,CMOS VCMOS/
2 + 0.6 VCMOS
+ 0.7 V–
Quiet on ROW/COL bits during NAP/PDN
entry tNPQ 4–tCYCLE Figure 47
Offset between read data and CC packets
(same device) tREADTOCC 12 – tCYCLE Figure 51
Offset between CC packet and read data
(same device) tCCSAMTOREAD 8–tCYCLE Figure 51
CTM/CFM stable before NAP/PDN exit tCE 2–tCYCLE Figure 48
CTM/CFM stable af ter NAP/PDN entry tCD 100 – tCYCLE Figure 47
ROW packet to COL packet ATTN
framing delay tFRM 7–tCYCLE Figure 46
Maximum time in NAP mode tNLIMIT – 10.0 µsFigure 45
Refresh interval tREF –32msFigure 50
Current control interval tCCTRL 34 tCYCLE 100 ms ms/tCY
CLE
Figure 51
Temperature control int erval tTEMP – 100 ms Figure 23
TCE command to TCAL command tTCEN 150 – tCYCLE Figure 23
TCAL command to quiet window tTCAL 22tCYCLE Figure 23
Quiet window (no read data) tTCQUIET 140 – tCYCLE Figure 23
RDRAM delay (no RSL operations
allowed) tPAUSE – 200.0 µs page 43
1) MSE/MS are fields of the SKIP register. For this combination (skip override) the tDCW parameter range is
effectively 0.0 to 0.0.
2) This parameter also applies to a -800 or -711 part when operated with tCYCLE =3.33ns.
3) This parameter also applies to a -800 part when operated with tCYCLE =2.81ns.
4) tS,MIN and tH,MIN for other tCYCLE values can be interpolated between or extrapolated from the timings at the
3 specified tCYCLE values.
5) With VIL,CMOS =0.5VCMOS – 0.6 V and VIH,CMOS =0.5VCMOS +0.6V
6) Effective hold becomes tH4’ = tH4 + [PDNXA × 64 × tSCYCLE +tPDNXB,MAX] – [PDNX × 256 × tSCYCLE]
if [PDNX × 256 × tSCYCLE] < [PDNXA × 64 × tSCYCLE + tPDNXB,MAX]. See Figure 48.
Ta ble 20 Tim ing Cond itions (cont’d)
Parameter Symbol Limit Values Unit F igure
min. max.
Data Sheet 71 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Table 21 El ectrical Characteristics
Parameter Symbol Li mit Values Unit
min. max.
VREF current @ VREF,MAX IREF –10 10 µA
RSL output high current @ (0 ≤ VOUT ≤ VDD)IOH –10 10 µA
RSL IOL current @ VOL = 0.9 V, VDD,MIN, TJ,MAX1) IALL 30.0 90.0 mA
RSL IOL current resolution step ∆IOL –2.0mA
Dynamic output impedance rOUT 150 – Ω
CMOS input leakage current
@ (0 ≤VI,CMOS ≤ VCMOS)II,CMOS – 10.0 10.0 µA
CMOS output voltage @ IOL,CMOS = 1.0 mA VOL,CMOS –0.3V
CMOS output high voltage
@ IOH,CMOS = – 0.25 mA VOH,CMOS VCMOS –0.3 – V
1) This measurement is made in manual current control mode; i.e. with all output device legs sinking current.
Table 22 Timing Characteristics
Parameter Symb ol Limit Values Unit Figure
min. max.
tQCTM-to-DQA/DQB output t ime
@ tCYCLE =3.33ns
CTM-to-DQA/DQB output t ime
@ tCYCLE =2.81ns
CTM-to-DQA/DQB output t ime
@ tCYCLE =2.50ns
– 0.3501),3)
– 0.3002),3)
– 0.2603)
+ 0.3501),3)
+ 0.3002),3)
+ 0.2603)
ns
ns
ns
Figure 55
Figure 55
Figure 55
tQR, tQF DQA/DQB output rise and fall tim es 0.2 0.45 ns Figure 55
tQ1 SCK(neg)-to-SIO0 delay
@ CLOAD,MAX = 20 pF (SD read data
valid).
–10nsFigure 58
tQR1, tQF1 SIOOUT rise/fall @ CLOAD,MAX = 20 pF – 5 ns Figure 58
tPROP1 SIO0-to-SIO1 or SIO1-to-SIO0 delay @
CLOAD,MAX = 20 pF
–10nsFigure 58
tNAPXA NAP exit delay - phase A – 50 ns Figure 48
tNAPXB NAP exit delay - phase B – 40 ns Figure 48
tPDNXA PDN exit delay - phase A – 4 µsFigure 48
tPDNXB PDN exit delay - phase B – 9000 tCYCLE Figure 48
tAS ATTN-to-STBY power st ate delay – 1 tCYCLE Figure 46
tSA STBY-to-ATTN power st ate delay – 0 tCYCLE Figure 46
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 72 6.00
RSL - Clocking
Figure 53 is a timing diagram which shows the detailed requirements for the RSL c lock signals on
the Channel.
The CTM and CTMN are differential clock inputs used for transmitting information on the DQA and
DQB, outputs. Most timing is measured relative to the points where they cross. The tCYCLE
parameter is measured from the falling CTM edge to the falling CTM edge. The tCL and tCH
parameters are m easured from falling to rising and rising to falling edges of CTM. The tCR and tCF
rise- and fall-time parameters are measured at the 20% and 80% points.
tASN ATTN/STBY-to-NAP power st ate delay –8 tCYCLE Figure 47
tASP ATTN/STBY-to-PDN power state delay – 8 tCYCLE Figure 47
1) This parameter also applies to a -800 or -711 part when operated with tCYCLE =3.33ns.
2) This parameter also applies to a -800 part when operated with tCYCLE =2.81ns.
3) tQ,MIN and tQ,MAX for other tCYCLE values can be interpolated between or extrapolated from the timings at the
3 specified tCYCLE values.
Table 22 Timing Characteristics (cont’d)
Parameter Symb ol Limit Values Unit Figure
min. max.
Data Sheet 73 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 53 RSL Timing - Clock Signals
The CFM and C FMN are differential c lock outputs used for receiving information on the D QA, DQB,
ROW and COL outputs. Most timing is measured relative to the points where they cros s. The tCYCLE
parameter is measured from the falling CFM edge to the falling CFM edge. The tCL and tCH
parameters are m easured from falling to rising and rising to falling edges of CFM. The tCR and tCF
rise- and fall-time parameters are measured at the 20% and 80% points.
The tTR parameter specifies the phase difference that may be tolerated with respect to the CTM and
CFM differential cloc k inputs (the CTM pair is always earlier).
SPT04239
20%
CIL
V
CIH
50%
80%
V
CFMN
CFM
TR
t
CR
V
X-
t
CM
V
X+
V
t
CR
t
CF CF
t
CL
t
CH
t
CYCLE
t
CTM
80%
50%
CIH
CIL
20%
V
V
CTMN
CL
t
CYCLE
t
X-
V
CR CR
t
CF
CH
t
X+
V
V
CM
t
CF
t t
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 74 6.00
RSL - Receive Timing
Figure 54 is a timing diagram which shows the detailed requirements for the RSL input signals on
the Channel.
The DQA, DQB, ROW, and COL signals are inputs which receive information transmitted by a
Direct RAC on the Channel. Each signal is s ampled twice per tCYCLE int erval. The set/hold window
of the sample points is tS/tH. The sample points are c entered at the 0% and 50% points of a cycle,
measured relative to the crossing points of the falling CFM clock edge. The s et and hold parameters
are measured at the VREF voltage point of the input transition.
The tDR and tDF rise- and fall-time parameters are measured at the 20% and 80% points of the input
transition.
Figure 54 RSL Timing - Data Signals for Receive
SPT04240
20%
DIL
V
DIH
80%
V
DQB
DR
t t
S
CFM
80%
50%
CIH
CIL
20%
V
V
CFMN
X-
V
X+
V
V
CM
ROW
COL
DQA
t
DF
t
H
t
S
H
t
t
CYCLE
0.5 x
Even Odd
REF
V
Data Sheet 75 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
RSL - Transmit Timing
Figure 55 is a timing diagram which shows the detailed requirements for the RSL output signals on
the Channel.
The DQA and DQB signals are outputs to transmit information that is received by a Direct RAC on
the Channel. Each signal is driven twice per tCYCLE interval. The beginning and end of the even
transmit window is at the 75% point of the previous c ycle and at the 25% point of t he current cycle.
The beginning and end of the odd t ransmit window is at the 25% point and at the 75% point of the
current cycle. These t ransmit points are meas ured relat ive to t he c rossing point s of t he f alling C TM
clock edge. The size of the actual transmit window is less than the ideal tCYCLE/2, as indicated by the
non-zero values of tQ,MIN and tQ,MAX. The tQ parameters are measured at the 50% voltage point of the
output transition.
The tQR and tQF rise- and fall-time parameters are measured at the 20% and 80% points of the
output transition.
Figure 55 RSL Timing - Data Signals for Transmit
CMOS - Receive Timing
Figure 56 is a timing diagram which shows the detailed requirements for the CM OS input signals.
The CMD and SIO0 signals are inputs w hich receive information transmitted by a c ontroller (or by
another RDRAM’s SIO1 output. SCK is the CMOS clock signal driven by the cont roller. All signals
are high true.
SPT04241
20%
QL
V
QH
50%
80%
V
DQB
QR
tt
Q, MAX
CTM
80%
50%
CIH
CIL
20%
V
V
CTMN
X-
V
X+
V
V
CM
DQA
t
QF
t
CYCLE
0.25 x
0.75 x
CYCLE
t
Q, MIN
t
Q, MAX
tt
Q, MIN
CYCLE
0.75 x
t
OddEven
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 76 6.00
The cycle time, high phase time, and low phase time of the SCK clock are tCYCLE1, tCH1 and tCL1, all
measured at the 50% level. The rise and fall times of SCK, CMD, and SIO0 are tDR1 and tDF1,
measured at the 20% and 80% lev els.
The CMD signal is sampled twice per tCYCLE1 interval, on the rising edge (odd data) and the falling
edge (even data). The set/hold window of the sample points is tS1/tH1. The SCK and CMD timing
points are measured at the 50% lev el.
The SIO0 s ignal is sampled once per tCYCLE1 interval on the falling edge. The set/hold window of the
sample points is tS2/tH2. The SCK and SIO0 timing points are measured at the 50% level.
Figure 56 CMOS Timing - Data Signals for Receive
SPT04242
20%
IL, CMOS
V
IH, CMOS
50%
80%
V
CMD
DR2
t t
S1
SCK
80%
50%
IH, CMOS
IL, CMOS
20%
V
V
t
DF2
t
H1
t
CYCLE1
50%
20%
V
SIO0
V
80%
IH, CMOS
IL, CMOS
t
DR2
CH1
t
CL1
tt
H1
t
S1
DF1
t
DR1
tt
t
S2
H2
Even Odd
DF2
t
Data Sheet 77 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
The SCK clock is also used for sampling data on RSL inputs in one situation. Figure 48 show s the
PDN and NAP exit sequences. If the PSX field of the INIT register is one (see Figure 27), then the
PDN and NAP exit sequences are broadcast; i.e. all R DRAMs that are in PDN or NAP will perform
the exit sequence. If the PSX field of the INIT register is zero, then the PDN and NAP exit sequences
are directed; i.e. only one RDRAM t hat is in PDN or NAP w ill perform the exit sequence.
The address of that RDRAM is specified on the DQA[5:0] bus in the set hold window tS3/tH3 around
the rising edge of SC K. This is shown in Figure 57. The SCK timing point is measured at the 50%
level, and the DQA[5:0] bus signals are measured at the VREF level.
Figure 57 CMOS Timing - Device Address for NAP or PDN Exit
CMOS - Transmit Timing
Figure 58 is a timing diagram which shows the detailed requirements for the CMOS output signals.
The SIO0 signal is driven once per tCYCLE1 interval on t he falling edge. The clock-to-output w indow
is tQ1,MIN/tQ1,MAX. The SCK and SIO0 timing points are measured at the 50% level. The rise and fall
times of SIO0 are tQR1 and tQF1, measured at the 20% and 80% levels.
SPT04243
20%
DIL
V
DIH
80%
V
DQA(5:0)
SCK
80%
50%
IH, CMOS
IL, CMOS
20%
V
V
t
H3
t
S3
PDEV
V
REF
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 78 6.00
Figure 58 CMOS Timing - Data Signals for Transmit
V
IH,CMOS
50%
V
IL,CMOS
80%
20%
SCK
SIO0
t
QR1
t
QF1
V
OH,CMOS
50%
V
OL,CMOS
80%
20%
t
Q1,MAX
V
IH,CMOS
50%
V
IL,CMOS
80%
20%
t
Q1,MIN
V
OH,CMOS
50%
V
OL,CMOS
80%
20%
SIO0
t
DR1
t
DF1
t
QR1
t
QF1
t
PROP1,MAX
t
PROP1,MIN
or
SIO1
SIO1
or
SIO0
Data Sheet 79 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 58 also shows the combinational path connecting SIO0 to SIO1 and the path connecting
SIO1 to SI O0 (read data only). The tPROP1 parameter specified this propagation delay . The rise and
fall times of SIO0 and SIO1 inputs must be tDR1 and tDF1, measured at the 20% and 80% levels. The
rise and fall times of SIO0 and SIO1 outputs are tQR1 and tQF1, measured at the 20% and 80% levels.
RS L - Dom a in Cross ing Win dow
When read data is returned by the RDRAM, information must cross from the receive clock domain
(CFM) to the transmit clock domain (CTM). The tTR param eter permits the CFM t o CTM phase to
vary through an entire cycle; i.e. there is no restriction on the alignment of these two clocks. A
second parameter tDCW is needed in order to describe how the delay between a RD command
packet and read data packet varies as a function of the tTR value.
Figure 59 shows this timing for five distinct values of tTR. Case A (tTR = 0) is what has been used
throughout this document. T he delay between the RD command and read data is tCAC. As tTR varies
from zero to tCYCLE (cases A through E), the command to data delay is (tCAC –tTR). When the tTR
value is in the range 0 to tDCW,MAX, the command to data delay can also be (tCAC –tTR –tCYCLE). This
is shown as cases A’ and B’ (the gray packets). Similarly, when the tTR value is in the range
(tCYCLE +tDCW,MIN) to tCYCLE, the command to data delay can also be (tCAC –tTR +tCYCLE). This is
shown as cases D’ and E’ (the gray pack ets). The RDRAM will work reliably with either the white or
gray packet tim ing. The delay value is selected at initialization, and remains fixed thereafter.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 80 6.00
Figure 59 RSL Transmit - C rossing Read Domains
SPA04245
CFM
CTM
CYCLE
t
TR
t
TR
t
= 0
= 0
TR
t
Case A
Case A’
CTM
TR
t
Case B’
Case B
TR
TR
t
t
=
=
DCW, max
t
t
DCW, max
Case C
CTM
t
TR CYCLETR
t
= 0.5
t
*
Case D
CTM
TR
t
=
TR
t
CYCLE
t
DCW, min
t
+
Case D’
DCW, minCYCLETR
t
=
t
+
t
Case E’
Case E
t
TR CYCLE
CYCLE
TR
TR
t
t
t
=
=
t
CTM
Q(a1)
RD a1
CAC
t
TR
t
-
tt
CAC
-
TR
t
+
CYCLE
t
CAC
t
-
TR
t
+
CAC
t
-
TR
t
CYCLE
Q(a1)
Q(a1)
CAC
t
-
TR
t
t
CAC
CAC
t
-
t
-
TR
TR
t
-
t
CYCLE
Q(a1)
-
CAC
t
CAC
t
Q(a1)
CYCLE
-
TR
tt
TR
t
-
Q(a1)
Q(a1)
Q(a1)
Q(a1)
COL
DQA/B
DQA/B
DQA/B
DQA/B
DQA/B
DQA/B
DQA/B
DQA/B
DQA/B
Data Sheet 81 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Timing Parameters
Table 23 Timing Parameter Summary
Parameter Description Min
-40
-800
Min
-45
-800
Min
-45
-711
Min
-53
-600
Max Unit Figure
tRC Row Cycle time of RDRAM banks - the
int erval bet we en ROWA packe ts wi th AC T
commands to the same bank.
28 28 28 28 –tCYCLE Figure 15
Figure 16
tRAS RAS-asserted ti me of RD RA M bank - the
i nterval between RO WA packet with AC T
command and ne xt ROWR packet with
PRER1) command to the s ame bank.
20 20 20 20 64 µs2) tCYCLE Figure 15
Figure 16
tRP Row Precharge time of RDRAM banks -
the interval between ROWR packet with
PRERa command and next RO WA packe t
with ACT command to the same bank.
8888–
tCYCLE Figure 15
Figure 16
tPP Precharge-to-precharge time of RDRAM
device - the interval between successive
ROWR packets with PRERa commands to
any banks of the same device.
8888–
tCYCLE Figure 12
tRR RAS-to-RAS t ime o f RDRAM device - the
interval between successive ROWA
packets with ACT commands to any
banks of the same device.
8888–
tCYCLE Figure 13
tRCD RAS-to-CAS Delay - the i nterval f rom
ROWA packet with ACT command to
COLC packet with RD or WR command).
Note - the RAS-to-CAS delay seen by the
RDRAM core (tRCD-C) i s equal to tRCD-C = 1
+ tRCD because of differences in the row
and column paths through the RDRAM
interface.
7977–
tCYCLE Figure 15
Figure 16
tCAC CAS Access delay - the interval from RD
comma nd to Q rea d d ata . The eq uation f or
tCAC is gi v e n in the TP ARM reg is ter in
Figure 39.
888812
tCYCLE Figure 4
Figure 39
tCWD CAS Writ e Delay (interval from WR
command to D write data. 66666tCYCLE Figure 4
tCC CAS-to-CAS time of RDRAM bank - the
interval between successive COLC
commands).
4444-
tCYCLE Figure 15
Figure 16
tPACKET Length of RO WA, ROWR, COLC, COLM
or COLX packet. 44444tCYCLE Figure 3
tRTR Interv a l fro m CO L C pa c k e t wit h W R
comma nd to COLC packet which causes
retire, and to COLM packet with
bytemask.
8888-
tCYCLE Figure 17
tOFFP The interval (offset) from COLC packet
with RDA command, or from COLC packet
wit h retir e command (aft er WRA automat ic
precharge), or from COLC packet wit h
PREC command, or from COLX packet
with PREX command to the equivalent
ROWR packet with PRER. The equation
for tOFFP is given in the TPARM register in
Figure 39.
44444
tCYCLE Figure 14
Figure 39
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 82 6.00
tRDP Inte rv a l fro m la st C OL C pa c ke t w it h RD
command to ROWR packet with PRER. 4444–tCYCLE Figure 15
tRTP Interv a l fro m la st C OL C pa c ke t w it h
automatic retire command to RO WR
packet with PRER.
4444–
tCYCLE Figure 16
1) Or equivalent PREC or PREX command. See Figure 14.
2) This is a constraint imposed by the core, and is therefore in units of µs rather than tCYCLE.
Table 24 Absolute Maximum Ratings
Parameter Symbol Limit Values Unit
min. max.
Voltage applied to any RSL or C MOS pin
with respect to GND VI,ABS –0.3 VDD + 0.3 V
Voltage on VDD and VDDA with respect to
GND VDD,ABS,
VDDA,ABS
–0.5 VDD +1.0 V
Storage temperature TSTORE – 50 100 °C
Table 23 Timing Parameter Summary (cont’d)
Parameter Description Min
-40
-800
Min
-45
-800
Min
-45
-711
Min
-53
-600
Max Unit Figure
Data Sheet 83 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
IDD - Supply Current Profile
Tab le 25 Su pply Curre nt P rof ile
RDRAM Blocks Co nsuming Power 1)
1) The CMOS interface consumes power in all power states.
IDD Value Limit Values Unit
max.
-800 max.
-711 max.
-600
Self-refresh only for INIT.LSR = 0 IDD,PDN 6000 6000 6000 µA
T/RCLK-Nap IDD,NAP 444mA
T/RCLK, ROW-demux IDD,STBY 100 95 90 mA
T/RCLK, ROW-demux, COL-demux IDD,ATTN 150 145 140 mA
T/RCLK, ROW-demux, COL-demux, DQ-
demux,
1 × WR-SenseAmp, 4 × ACT-Bank
IDD,ATTN-W 635 570 495 mA
T/RCLK, ROW-demux, COL-demux, DQ-
mux,
1 × RD-SenseAmp, 4 × ACT-Bank 2)
2) This does not includ e the IOL si nk current. The RDRAM dis sipates IOL ×VOL i n each output drive r when a logic
one is driven.
IDD,ATTN-R 5752) 520 450 mA
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 84 6.00
Capacitance and Inductance
Figure 60 shows t he equivalent load circuit of the RSL and CM OS pins. The circuit models the load
that the dev ice presents to the Channel.
Figure 60 Equivalent Load Circuit for RSL Pins
SPT04246
Ι
R
Ι
C
L
Ι
Pad DQA, DQB, RQ Pin
Gnd Pin
Pad
C
R
Ι
Ι
L
Ι
Gnd Pin
CTM, CTMN,
Pad
C
Ι
L
Ι
Gnd Pin
SCK, CMD Pin
Pad
C
Ι
Gnd Pin
SIO0, SIO1 Pin
CFM, CFMN Pin
, CMOS
, CMOS
, CMOS
L
Ι
, CMOS, SIO
Data Sheet 85 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
This circuit does not include pin coupling effects that are often present in the packaged device.
Because coupling effects make the effective single-pin inductance LI, and capacitance CI, a
function of neighboring pins, these parameters are intrinsically data-dependent. For purposes of
specifying the device electrical loading on the Channel, the effective LI and CI are defined as the
worst-case values ov er all specified operating conditions.
LI is defined as the effective pin induc tance based on the device pin assignment. Because the pad
assignment places each RSL signal adjacent to an AC ground (a GND or VDD pin), the effective
inductance must be defined based on this configuration. Therefore, LI as sumes a loop with the RSL
pin adjacent to an AC ground.
CI is defined as the effective pin c apacitance based on t he dev ice pin assignment. I t is the sum of
the effective pack age pin capacitance and the IO pad capacitance.
Table 26 R SL Pin Parasitics
Parameter and Conditions - RSL Pins Symbol Limit Values Unit
min. max.
RSL effective input inductance LI –4.0 nH
Mutual inductance bet ween any DQA or DQB
RSL signals. L12 –0.2 nH
Mutual inductance between any ROW or C OL
RSL signals. –0.6 nH
Difference in LI value between any RSL pins of
a single device. ∆LI –1.8 nH
RSL effective input capacitance1) -800
RSL effective input capacitance1)-711
RSL effective input capacitance1)-600
1) This value is a combination of the device IO circuitry and package capacitances.
CI 2.0
2.0
2.0
2.4
2.4
2.6
pF
pF
pF
Mutual capacitance between any RSL signals. C12 –0.1 pF
Difference in CI value between average of
CTM/CFM and any RSL pins of a single device. ∆CI –0.06 pF
RSL effective input resistance RI 415 Ω
Table 27 C MOS Pin Parasitics
Parameter and Conditions - CMOS Pins Symbol Limit Values Unit
min. max.
CMOS effective input inductance LI ,CMOS –8.0nH
CMOS effective input capacitance (SCK,CMD)1)
1) This value is a combination of the device IO circuitry and package capacitances.
CI ,C M O S 1.7 2.1 pF
CMOS effective input capacitance (SIO1, SIO0)1) CI ,CMO S, SIO –7.0pF
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 86 6.00
Center-Bonded FBGA Package
Figure 61 shows the form and dimensions of the recommended package for the center-bonded CSP
device class.
Figure 61 Center-Bonded FBGA Package
Table 28 lists the numerical values corresponding to dimensions shown in Figure 61.
ABCDEFGHJ
1
2
3
4
5
6
7
D
A
e1
d
E
E1
8e2
TopBottom
Bottom
Bottom
9
10
KLMN
Data Sheet 87 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Table 28 Center-Bond ed FBGA Package Dimensions
Parameter Symbol Limit Values Unit
min. max.
Ball pitch (x-axis) e1 1.00 1.00 mm
Ball pitch (y-axis) e2 0.8 0.8 mm
Package body length A 10.9 11.1 mm
Package body width D 10.4 10.6 mm
Package total thickness E 0. 65 1.05 mm
Ball height E1 0.18 0.35 mm
Ball diameter d 0.3 0.4 mm
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 88 6.00
Int er le aved Devic e Mo de
Interleaved Device Mode permits a group of eight RDRAMs on the Channel to collectively respond
to a command. The purpose of this collective response is to limit the number of bits in eac h dualoct
data packet which are read from or written to a single RDRAM device. This capability permits a
memory controller to implement hardware for fault detection and correction that can t olerate the
complete internal failure of one RDRAM device on a Channel.
The IDM bit of the INIT control register enables this fault tolerant operating mode. When it is set,
the RDRAM w ill interpret the DR4..0 and DC4..0 fields of the ROW and COLC packets differently.
Figure 63 shows the differences using an ex ample system with ei ght RDRAMs.
The DEVID4..0 regist ers of these RDRAMs are initialized to “00000” through “00111’. However,
when the ID M bit is set, only the upper t wo bits (DEVID4..3) will be c ompared to the DR4..3 and
DC4..3 fields. This means that ROW and COLC packets will be ex ecuted by groups of eight
RDRAMs, with a Channel containing from one to four of these group s. The low-order DR2..0 bits
are not used when IDM is set , and the low-order DC2..0 bits have a modified function described
below.
With IDM set, a directed ACT or PRE command in a ROW packet causes eight RDR AMs to
perform the indic ated operation. Likewise, when a R D or WR command is specified in a COLC
command, the s elected group of eight RDRAMs responds. When using IDM, devices must be
added to t he Channel in groups of eight. An application will typically make the IDM bit setting the
same for all RDRAMs on a C hannel.
The mechanism for indicating a broadcast ROW packet (DR4F and DR4T are both set to one) is
not affec ted by the setting of the IDM bit; i.e. IDM mode does not change the broadcast ROW
packet mechanism.
Likewise, the C OLX fields (DX4..0, XOP4..0, and BX5..0) are not changed by IDM mode - all
COLX pac kets are directed t o a single device.
When the ID M bit is set, COLM packets should not be used (the M bit should be set to zero,
selecting only COLX packets). This is because the mapping of bytes to RDRAM storage cells is
changed by ID M mode.
Returning to Figure 63, the remaining fields of the ROW and COLC packets are interpret ed in the
same way regardless of the setting of the IDM bit - IDM mode does not affect these fields. Specifi-
cally, the BR5..0 and BC5..0 fields of the ROW and COLC packets are used to select one of the
banks just as when IDM is not set. The R8..0 field of the R OW packet selects a row of the selected
(BR5..0) bank to load into the bank’s sense amp. And the C6..0 field selects one dualoc t of the
selected (BC5..0) bank’s sense amp.
The IDM bit aff ects what is done with this selected dualoct . When IDM is not set, the dualoct is
driven onto the Channel by t he single selected RDRAM device. When IDM is s et, each RDRAM of
the eight devic e group selected by DC4..3 drives 16 or 24 bits (x18 device) of the 144-bit dualoct.
The bits driven are a function of the DEVID2..0 RDRAM register field, the DC2..0 COLC packet
field, and t he dev ice width ( x18). Figure 63 shows the mapping that is appropriat e f or D C 2..0=000.
Figure 64 and Figure 65 show the mapping for all eight values of DC2..0. There are eight
mappings, which are rotated among the eight devices using the following equation:
Data Sheet 89 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Pin = 7 - 4•(DEVID2^DC2)
- 2•(DEVID1^DC1) - 1•(DEVID0^DC0) (Eq 1)
where “^” is the exclusive-or function. “Pin” is the pin number that is dri ven by the RDRAM with the
DEVID2..0 value. For example, Pin=0 means the RDRAM drives DQA0 and DQB0, and so forth.
The DQA8 pin is always driven with DQA7, and DQB8 is always driven with DQB6 for x18 devices.
For each of the eight mappings, the eight-RDRAM group supplies a complete dualoct. As the
application steps through eight values of DC2..0, all the bit s of the eight underlying dualocts will be
accessed. Thus, an eight-RDRAM group appears to be a single RDRAM with eight times the
normal page size, with the DC2..0 field providing the extra column addressing information (beyond
what C6..0 provides).
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 90 6.00
Figure 62: ACT, PRE, RD and WR Commands for Eig ht RDRAM System wi th IDM=1
DQA0
DQB0
DQA1
DQB1
DQA2
DQB2
DQA3
DQB3
DQA4
DQB4
DQA5
DQB5
DQA6
DQB6
DQA7
DQB7 DQB8DQA8
••• •••
C 6 ..0
DC2 ..0
DC4 ..3
dualoct
form
column
access
•••
R 1 2 ..0
DR4 ..3
row
access
••••••
••••••
••••••
•••
one bank
RD
WR
ACTPRE
device
access
BC5..0
BR5..0
bank
access
DEVID4..3
com pare to
•••
•••
•••
•••
same a s
device 0
same a s
device 0 sae as
device 0
sam e as
device 0 same as
device 0 same as
device 0
same as
device 0
bank array
sense
amp
0000 0
DEVID
4..0 0000 1 0001 0 00 011 00100 00101 00110 00 1 1 1
CTM/CFM
DQA8
DQB0
DQA0
DQA8
••••••
Channel
d ua loct (144 b i ts ) one bit
••• •••
row (2
C
du alocts)bank (2
R
ro ws)
•••
••• •••
••• •••
••• •••
•••
•••
•••
•••
•••
device (2
B
banks)
notation
RDRAM 0 RDRAM 1 RDRAM 2 RDRAM 3 RDRAM 4 RDRAM 5 RDRAM 6 RDRAM 7
= 000
Data Sheet 91 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Figure 63: M apping from DEVID2..0 and DC2..0 Fields to DQ Packet wi th IDM=1
100 101 110 11 1 000 001 010 011
DC2..0
CTM/CFM
DEVID2..0
Mapping for
previous figure
DQA0
DQA1
DQA2
DQA3
DQA4
DQA5
DQA6
DQA7
DQA8
DQB0
DQB1
DQB2
DQB3
DQB4
DQB5
DQB6
DQB7
DQB8
000
DQA0
DQB0
DQA1
DQB1
DQA2
DQB2
DQ3
DQB3
DQA4
DQB4
DQA5
DQB5
DQA6
DQB6
DQA7
DQB7 DQB8DQA8
001
DQA0
DQB0 DQA1
DQB1
DQA2
DQB2 DQA3
DQB3
DQA4
DQB4 DQA5
DQB5
DQA6
DQB6 DQA7
DQB7
DQB8 DQA8
010
DQA0
DQB0
DQA1
DQB1 DQA2
DQB2
DQA3
DQB3
DQA4
DQB4
DQA5
DQB5 DQA6
DQB6
DQA7
DQB7 DQB8DQA8
011
DQA0
DQB0 DQA1
DQB1 DQA2
DQB2 DQA3
DQB3
DQA4
DQB4 DQA5
DQB5 DQA6
DQB6 DQA7
DQB7
DQB8 DQA8
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 92 6.00
Figure 65: M apping from DEVID2..0 and DC2..0 Fields to DQ Packet wi th IDM=1 (continued)
100 101 110 11 1 000 001 010 011
DC2..0
CTM/CFM
DEVID2..0
DQA0
DQA1
DQA2
DQA3
DQA4
DQA5
DQA6
DQA7
DQA8
DQB0
DQB1
DQB2
DQB3
DQB4
DQB5
DQB6
DQB7
DQB8
100
DQA0
DQB0
DQA1
DQB1
DQA2
DQB2
DQA3
DQB3 DQA4
DQB4
DQA5
DQB5
DQA6
DQB6
DQA7
DQB7 DQB8DQA8
101
DQA0
DQB0 DQA1
DQB1
DQA2
DQB2 DQA3
DQB3 DQA4
DQB4 DQA5
DQB5
DQA6
DQB6 DQA7
DQB7
DQB8 DQA8
110
DQA0
DQB0
DQA1
DQB1 DQA2
DQB2
DQA3
DQB3 DQA4
DQB4
DQA5
DQB5 DQA6
DQB6
DQA7
DQB7 DQB8DQA8
111
DQA0
DQB0 DQA1
DQB1 DQA2
DQB2 DQA3
DQB3 DQA4
DQB4 DQA5
DQB5 DQA6
DQB6 DQA7
DQB7
DQB8 DQA8
Data Sheet 93 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
Glossary of Terms
ACT A ctivate command from AV field.
activate To access a row and place in sense amp.
adjacent Two RDRAM banks which share sense amps (also called doubled banks).
ASYM CCA register field for RSL VOL/VOH.
ATTN Power state - ready for ROW/COL packets.
ATTNR Power state - transmitting Q packets.
ATTNW Power state - receiving D packets.
AV Opcode field in ROW packets.
bank A block of 2RBIT × 2CBIT storage cells in the core of the RDRAM.
BC Bank address field in COLC packet.
BBIT CNFGA register field - # bank address bits.
broadcast An operation executed by all RDRAMs.
BR Bank address field in ROW packets.
bubble Idle cycle(s) on RDRAM pins needed because of a resource constraint.
BYT CNFGB register field - 8/9 bits per byte.
BX Bank address field in COLX packet.
CColumn address field in COLC packet.
CAL Calibrate (IOL) command in XOP field.
CBIT CNFGB register field - # column address bits.
CCA Control register - current control A.
CCB Control register - current control B.
CFM,CFMN Clock pins for receiving packets.
Channel ROW/COL/DQ pins and external wires.
CLRR Clear reset command from SOP field.
CMD CMOS pin for initialization/power control.
CNFGA Control register with configuration fields.
CNFGB Control register with configuration fields.
COL Pins for column-access control.
COL COLC,COLM,COLX packet on COL pins.
COLC Column operation packet on COL pins.
COLM Write mask packet on COL pins.
column Rows in a bank or activated row in sense amps have 2CBIT dualocts column storage.
command A decoded bit-combination from a field.
COLX Extended operation packet on COL pins.
controller A logic-device which drives the ROW/COL /DQ wires for a Channel of RDRAMs.
COP Column opcode field in COLC packet.
core The banks and sense amps of an RDRAM.
CTM,CTMN Clock pins for transmitting packets.
current control Periodic operations to update the proper IOL value of RSL output drivers.
DWrite data packet on DQ pins.
DBL C NFGB register field - doubled-bank.
DC Device address field in COLC packet.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 94 6.00
device An RDRAM on a Channel.
DEVID Control register with device address that is matched against DR, DC, and DX fields.
DM Device match for ROW packet decode.
doubled-bank RDRAM with shared sense amp.
DQ DQA and DQB pins.
DQA Pins for data byte A.
DQB Pins for data byte B.
DQS NAPX register field - PDN/NAP exit.
DR,DR4T,DR4F Device address field and packet framing fields in ROWA and ROWR packets.
dualoct 16 bytes - the smallest addressable datum.
DX Device address field in COLX packet.
field A collection of bits in a packet.
INIT Control register with initialization fields.
initialization Configuring a Channel of RDRAMs so they are ready to respond to transactions.
LSR CNFGA register field - low-power self-refresh.
MMask opcode field (COLM/COLX packet).
MA Field in COLM packet for masking byte A.
MB Field in COLM packet for masking byte B.
MSK Mask command in M field.
MVER Control register - manufacturer ID.
NAP Power state - needs SCK/CMD wakeup.
NAPR N ap command in ROP field.
NAPRC Conditional nap command in ROP field.
NAPXA NAPX register field - NAP exit delay A.
NAPXB NAPX register field - NAP exit delay B.
NOCOP No-operation command in COP field.
NOROP N o-operation command in ROP field.
NOXOP N o-operation command in XOP field.
NSR INIT register field- NAP self-refresh.
packet A collection of bits carried on the Channel.
PDN Power state - needs SCK/CMD wakeup.
PDNR Powerdown command in ROP field.
PDNXA Control register - PDN exit delay A.
PDNXB Control register - PDN exit delay B.
pin efficiency The fraction of non-idle cycles on a pin.
PRE PREC,P RER,PREX precharge commands.
PREC Precharge command in COP field.
precharge Prepares sense amp and bank for activate.
PRER Precharge command in ROP field.
PREX Precharge command in XOP field.
PSX INIT register field - PDN/NAP exit.
PSR INIT register field - PDN self-refresh.
PVER C NFGB register field - protocol version.
Data Sheet 95 6.00
Direct RDRAM
288- M bi t (512kx18x32s)
QRead data packet on DQ pins.
RRow address field of ROWA packet.
RBIT CNFGB register field - # r ow addr ess bits.
RD/RDA Read (/precharge) command in COP field.
read Operation of accessing sense amp data.
receive Moving information from the Channel into the RDRAM (a serial stream is demuxed).
REFA Refresh-activate command in ROP field.
REFB Control register - next bank (self-refresh).
REFBIT CNFGA register field - ignore bank bits (for REFA and self-refresh).
REFP Refresh-precharge command in ROP field.
REFR Control register - next row for REFA.
refresh Periodic operations to restore storage cells.
retire The automatic operation that stores write buffer into sense amp after WR command.
RLX RLXC,RLXR,RLXX relax commands.
RLXC Relax command in COP field.
RLXR Relax command in ROP field.
RLXX Relax command in XOP field.
ROP Row-opcode field in ROWR packet.
row 2CBIT dualocts of cells (bank/sense amp).
ROW Pins for row-access control
ROW ROWA or ROWR packets on ROW pins.
ROWA Activate packet on ROW pins.
ROWR Row operation packet on ROW pins.
RQ Alternate name for ROW/COL pins.
RSL Rambus Signaling Levels.
SAM Sample (IOL) command in XOP field.
SA Serial address packet for control register transactions w/ SA address field.
SBC Serial broadcast field in SRQ.
SCK CMOS clock pin.
SD Serial data packet for control register transactions w/ SD data field.
SDEV Serial device address in SRQ packet.
SDEVID IN IT register field - Serial device ID.
self-refresh Refresh mode for PDN and NAP.
sense amp Fast storage that holds copy of bank’s row.
SETF Set fast clock command from SOP field.
SETR Set reset command from SOP field.
SINT Serial interval packet for control register read/write transactions.
SIO0,SIO1 CMOS serial pins for control registers.
SOP Serial opcode field in SRQ.
SRD Serial read opcode command from SOP.
SRP INIT register field - Serial repeat bit.
SRQ Serial request packet for control register read/write transactions.
STBY Power state - ready for ROW packets.
Direct RDRAM
288- M bi t (512kx18x32s)
Data Sheet 96 6.00
SVER Control register - stepping version.
SWR Serial write opcode command from SOP.
TCAS TCLSCAS register field - tCAS core delay.
TCLS TCLSCAS register field - tCLS core delay.
TCLSCAS Control register - tCAS and tCLS delays.
TCYCLE Control register - tCYCLE delay.
TDAC Control register - tDAC delay.
TEST77 Control register - for test purposes.
TEST78 Control register - for test purposes.
TRDLY Control register - tRDLY delay.
transaction ROW,COL,DQ packets for memor y access.
transmit Moving information from the RDRAM onto the Channel (parallel word is muxed).
WR/WRA W rite (/precharge) command in COP field.
write Operation of modifying sense amp data.
XOP Extended opcode field in COLX packet
INFINEON Technologies
Att e n tio n pl e ase !
As far as patents or other rights of third parties are concerned, liability is only
assumed for components, not for applications, processes and c ircuits implemented
within components or assemblies. This infomation describes the type of
components and shall not be considered as assured characteristics. Terms of
delivery and rights to change design reserved.
For questions on technology, delivery and prices please contact INFINEON
Technologies Offices in Munich or the INFINEON Tec hnologies Sales Offices and
Representatives worldwide.
Due to technical requirements components may contain dangerous substances.
For information on the types in question please contact your nearest INFINEON
Technologies office or representative.
Packing
Please use the recycling operators known to you. We can help you - get in touch
with your nearest sales office. By agr eement we will take packing material back, if
it is sorted. You must bear the costs of transport. For packing material that is
returned to us unsorted or which we are not obliged to accept, we shall have to
invoice you for any costs incurred.
Components used in life-support devices or systems must be
expressly authorized for such purpose!
Ciritcal components1 of INFINEON Technologies, may only be used in life-support
devices or s yst ems2 with the ex press written app rov al of INFINEON Technol ogies.
1. A critical component is a component used in a life-support device or system
whose failure can reasonably be expected to c ause the failure of that life-support
device or system, or to affect its safety or effectiveness of that device or system.
2. Life support devices or systems are intended (a) to be implanted in the human
body, or (b) to support and/or maintain and sustain human life. If they fail, it is
reasonable to assume that the health of the user may be endangered.
