GS8672T18/36BE-400/333/300/250/200
165-Bump BGA
Commercial Temp
Industrial Temp
Features
• On-Chip ECC with virtually zero SER
• Simultaneous Read and Write SigmaDDR™ Interface
• Common I/O bus
• JEDEC-standard pinout and package
• Double Data Rate interface
• Byte Write Capability
• Burst of 2 Read and Write
• 1.8 V +100/–100 mV core power supply
• 1.5 V or 1.8 V HSTL Interface
• Pipelined read operation with self-timed Late Write
• Fully coherent read and write pipelines
• ZQ pin for programmable output drive strength
• IEEE 1149.1 JTAG-compliant Boundary Scan
• Pin-compatible with 18Mb, 36Mb and 144Mb devices
• 165-bump, 15 mm x 17 mm, 1 mm bump pitch BGA package
• RoHS-compliant 165-bump BGA package available
72Mb SigmaDDR-II™
Burst of 2 ECCRAM™
400 MHz–200 MHz
1.8 V V
DD
1.8 V and 1.5 V
DDQ
C clock inputs. C and C are also independent single-ended
clock inputs, not differential inputs. If the C clocks are tied
high, the K clocks are routed internally to fire the output
registers instead.
Each internal read and write operation in a SigmaDDR-II B2
ECCRAM is two times wider than the device I/O bus. An input
data bus de-multiplexer is used to accumulate incoming data
before it is simultaneously written to the memory array. An
output data multiplexer is used to capture the data produced
from a single memory array read and then route it to the
appropriate output drivers as needed.
When a new address is loaded into the part, A0 is used to
initialize the pointers that control the data multiplexer / de-
multiplexer so the ECCRAM can perform "critical word first"
operations. From an external address point of view, regardless
of the starting point, the data transfers always follow the same
sequence {0, 1} or {1, 0} (where the digits shown represent
A0).
SigmaDDR™ ECCRAM Overview
The GS8672T18/36BE SigmaDDR-II ECCRAMs are built in
compliance with the SigmaDDR-II SRAM pinout standard for
Common I/O synchronous SRAMs. They are 75,497,472-bit
(72Mb) SRAMs. The GS8672T18/36BE SigmaDDR-II
SRAMs are just one element in a family of low power, low
voltage HSTL I/O SRAMs designed to operate at the speeds
needed to implement economical high performance
networking systems.
On-Chip Error Correction Code
GSI's ECCRAMs implement an ECC algorithm that detects
and corrects all single-bit memory errors, including those
induced by Soft Error Rate (SER) events such as cosmic rays,
alpha particles etc. The resulting SER of these devices is
anticipated to be <0.002 FITs/Mb — a 5-order-of-magnitude
improvement over comparable SRAMs with no On-Chip ECC,
which typically have an SER of 200 FITs/Mb or more. SER
quoted above is based on reading taken at sea level.
However, the On-Chip Error Correction (ECC) will be
disabled if a “Half Write” operation is initiated. See the
Byte
Write Contol
section for further information.
Clocking and Addressing Schemes
The GS8672T18/36BE SigmaDDR-II SRAMs are
synchronous devices. They employ two input register clock
inputs, K and K. K and K are independent single-ended clock
inputs, not differential inputs to a single differential clock input
buffer. The device also allows the user to manipulate the
output register clock inputs quasi independently with the C and
Parameter Synopsis
-400
tKHKH
tKHQV
2.5 ns
0.45 ns
-333
3.0 ns
0.45 ns
-300
3.3 ns
0.45 ns
-250
4.0 ns
0.45 ns
-200
5.0 ns
0.45 ns
Rev: 1.02 1/2013
1/29
© 2011, GSI Technology
Specifications cited are subject to change without notice. For latest documentation see http://www.gsitechnology.com.
GS8672T18/36BE-400/333/300/250/200
Background
Common I/O SRAMs, from a system architecture point of view, are attractive in read dominated or block transfer applications.
Therefore, the SigmaDDR-II ECCRAM interface and truth table are optimized for burst reads and writes. Common I/O SRAMs
are unpopular in applications where alternating reads and writes are needed because bus turnaround delays can cut high speed
Common I/O SRAM data bandwidth in half.
Burst Operations
Read and write operations are "burst" operations. In every case where a read or write command is accepted by the ECCRAM, it
will respond by issuing or accepting two beats of data, executing a data transfer on subsequent rising edges of K and K, as
illustrated in the timing diagrams. This means that it is possible to load new addresses every K clock cycle. Addresses can be
loaded less often, if intervening deselect cycles are inserted.
Deselect Cycles
Chip Deselect commands are pipelined to the same degree as read commands. This means that if a deselect command is applied to
the ECCRAM on the next cycle after a read command captured by the ECCRAM, the device will complete the two beat read data
transfer and then execute the deselect command, returning the output drivers to high-Z. A high on the LD pin prevents the RAM
from loading read or write command inputs and puts the RAM into deselect mode as soon as it completes all outstanding burst
transfer operations.
SigmaDDR-II ECCRAM Read Cycles
The SRAM executes pipelined reads. The status of the Address, LD and R/W pins are evaluated on the rising edge of K. The read
command (LD low and R/W high) is clocked into the SRAM by a rising edge of K. After the next rising edge of K, the SRAM
produces data out in response to the next rising edge of C (or the next rising edge of K, if C and C are tied high). The second beat
of data is transferred on the next rising edge of C, for a total of two transfers per address load.
SigmaDDR-II ECCRAM Write Cycles
The status of the Address, LD and R/W pins are evaluated on the rising edge of K. The ECCRAM executes "late write" data
transfers. Data in is due at the device inputs on the rising edge of K following the rising edge of K clock used to clock in the write
command (LD and R/W low) and the write address. To complete the remaining beat of the burst of two write transfer, the
ECCRAM captures data in on the next rising edge of K, for a total of two transfers per address load.
Rev: 1.02 1/2013
5/29
© 2011, GSI Technology
Specifications cited are subject to change without notice. For latest documentation see http://www.gsitechnology.com.