Professional Documents
Culture Documents
WARANGAL
LPVLSI Assignment 03
Different designs of low power SRAM cells with their analysis
of Noise margin and Power Dissipation
Submitted by
Anish Kumar Mishra
Reg No:23ECM2R01
M.Tech (VLSI Design-1ST Year)
SRAM (Static Random Access Memory):
SRAM has emerged as a critical component in numerous VLSI chips owing to its
high storage density and swift access times. The significant advancements in
low-power, low-voltage memory design in recent years have spurred extensive
research into SRAM technology. This surge is largely fuelled by the escalating
demand for notebooks, laptops, IC memory cards, and handheld communication
devices. SRAMs find extensive application in mobile devices, serving as both
on-chip and off-chip memories, thanks to their user-friendly nature and minimal
standby leakage.
6T static random-access memory is a type of semiconductor memory that uses
bistable latching circuitry to store each bit. The term static differentiates it from
dynamic RAM which must be periodically refreshed. SRAM exhibits data
remembrance, but is still volatile in conventional sense, that data is eventually
lost when memory is not powered.
Schematic:
Sizing ratios of CMOS 6Ts cells:
Simulation Waveform:
Average Power = 221mW
7T SRAM Cell:
Operation:
Read Operation:
Schematic:
9T SRAM CELL
Operation:
Read Operation
The proposed cell's read operation mirrors that of the reference 8T cell, utilizing
transistors M7 and M8 to form a read buffer that separates the readout path from
the internal storage. Node "X" controls M7, allowing it to turn off when a '1' is
stored and conduct when a '0' is stored. During a read operation, the read bit line
(RBL) is precharged, and the read word line (RWL) is activated. If a '0' is stored,
the RBL discharges through M7 and M8, while if a '1' is stored, M7 blocks the
discharge path, maintaining the precharged level of RBL. A single-ended sensing
scheme determines whether RBL has discharged. Despite the reduced voltage
level when holding a '1', node "X" is always clamped to VDD or GND, ensuring
strong conduction through M7 (or M8) and maintaining equivalent read
performance to the 8T reference cell.
This decoupling of the readout path provides a read margin equivalent to the hold
margin, sufficient across various supply voltages and process variations.
However, read performance depends on factors like bitline capacitance, sense
amplifier sensitivity, and the drive strength of read buffer transistors (M7, M8).
Thus, read performance is tailored to specific applications and architectures, but
it significantly degrades with reduced supply voltage.
Write Operation
The SF-SRAM cell's intriguing aspect lies in its unique write operation
performance. Similar to an 8T cell, writing begins by setting the write bit lines
(WBL and WBLB) to the data level and activating the write word line (WWL).
In standard 8T cells, ensuring successful writes involves overcoming the pull-up
pMOS by the pull-down path. However, fluctuating currents due to process
variations often disrupt this balance, leading to failed writes. The proposed cell
employs a feedback loop from node "X" to weaken the pull-up path,
addressingthis issue and ensuring successful writes, especially in asymmetric
write scenarios.
Write 1 Operation:
When transitioning the cell from a hold '0' state to writing a '1', WBL is set to
VDD and WBLB is grounded. Activation of WWL initiates the write operation.
Initially, M9 facilitates contention to the pull-down path through M5, aiming to
pull towards an intermediate voltage between VDD and GND. In challenging
conditions like the SF corner, M9's feedback weakens as the cell is charged,
aiding in achieving a successful write by facilitating an "easy" write process. The
proposed SF-SRAM cell demonstrates significant advantages over the reference
8T cell, particularly evident at lower supply voltages, maintaining positive write
margins down to below 200 mV under global variations, whereas the 8T cell's
write margin becomes negative below 500 mV
Write 0 Operation:
During the Write '0' operation, assuming that node "X" is charged toVDD and
node "Y" is discharged to GND, WBL is grounded while WBLB is set to VDD.
Upon asserting WWL, node "X" begins at a lower voltage, and M9 is initially cut
off. This configuration ensures a strong pull-up current from M2 overwhelms any
residual pull-up current from M9, initiating rapid positive feedback in the cross-
coupled inverters to pull the storage nodes to their respective rails. This
straightforward process leverages the initial conditions and the cutoff state of M9
to achieve efficient write '0' operations.
Schematic:
Simulation Waveform