A Proposed Five Transistor CMOS SRAM Cell For High Speed Applications

A Proposed Five Transistor CMOS SRAM Cell For High Speed Applications

Dachineni Bharath Satyanarayana*, Kakarlapudi Pradeep*, Tsr Prasad**, T Ravi

*Electronics and Communication Engineering, K L University

** Asst. professor, Electronics and Communication Engineering, K L University

AbstractStatic random access memories (SRAMs) comprise an increasingly large portion of modern very large scale integrated (VLSI) circuits. A SRAM cell must meet stringent requirements for operation in the deep sub-micron ranges. This paper presents a new five transistor (5T) CMOS SRAM cell to accomplish improvements in stability, power dissipation, and performance over previous designs, for high speed and high stability memory operation. Designed circuit is simulated in a proprietary 180 nm CMOS process, using Cadence Spectre and BSIM3v3 models. Simulated results reveal that the proposed 5T SRAM cell exhibits considerable improvement in performance, power dissipation, and stability over both the conventional 6T and existing 5T SRAM cells without any area overhead.

Keywords CMOS, SRAM, VLSI, Static Noise Margin (SNM).

1. INTRODUCTION

   Colossal advances in CMOS technology have made it possible to design chips with high integration density, better performance, and low power consumption. To attain these objectives, the feature size of the CMOS devices has faced aggressive scaling down to very small features and dimensions. However, the leakage current has increased immensely with technology scaling, and has become a major contributor to the total IC power [1]. In addition, as feature size of CMOS devices scales down, the random variations in process parameters have emerged as a major design challenge in circuit design [2]. These random variations of device parameters in nano-scale CMOS technologies include random variations in channel length, channel width, oxide thickness, threshold voltage, etc [2]. These random parameter variations result in significant variation in the characteristic of digital circuits.

   Modern microprocessors employ on-chip caches, which can effectively reduces the speed gap between the processor and main memory to boost system performance. These on- chip caches are usually implemented using arrays of SRAM cells. A six transistor (6T) SRAM cell, shown in Fig. 1, is conventionally used as the memory cell. However, the mismatch in the strength between transistors of 6T SRAM cell due to process variations can results in failure during read operation, i.e. flipping of the cell data while reading, especially at lower levels of VDD [3]. Therefore, conventional SRAM cell shows poor stability at very small feature size. In addition, as CMOS technology scales down, an increase in total leakage current of a chip is observed.

   Fig. 1. Conventional 6T SRAM Cell

   Moreover, the total leakage current of chip is proportional to the number of transistors on the chip. Since the SRAM includes large number of transistors on a chip, the SRAM leakage has also become a more significant component of total chip leakage in scaled CMOS technology. Hence, stability during read operation and leakage current of SRAM cell are two most prominent parameters in designing of SRAM cell in nano-scale CMOS technologies.

   A novel 5T SRAM cell [4] has been previously proposed as an improvement to the standard six transistor (6T) SRAM cell model in various aspects. This 5T SRAM cell, as shown in Fig. 2, comprises two inverters, connected back-to-back and one additional transistor that is used to access the cell for read and write. Here both the bit-lines are precharged to VDD to retain the data during standby mode. However, the speed of a cell, which characterizes the performance of the cell, is still to be improved to reduce the speed gap further between processor and main memory.

   In response to these challenges in both conventional 6T and novel 5T SRAM cells, a new 5T SRAM cell has been proposed and its performance issues are discussed. The rest of the paper is organized as follows. Section II discusses basic structure, write, and read operations of the proposed 5T SRAM cell. Section III explores detailed static noise margin (SNM) analysis of the proposed cell under various modes. Section IV presents the simulation results performed

   Fig. 2. Novel 5T SRAM Cell

   in a proprietary 180 nm CMOS process. On the basis of results analyzed, section V concludes the significance of the proposed cell in various high speed, high stability, and low power applications.

2. PROPOSED 5T SRAM CELL

   Fig. 3 shows the proposed five transistor (5T) SRAM cell. In this cell, Inverter NMOS transistors (M1, M3) are directly connected to the bit lines, PMOS transistors (M2, M4) are connected to power supply voltage (VDD), and there is an additional transistor M5 coupling the inverters. Unlike standard cell, no word line transistors are needed to provide access during the read and write cycles. In contrast to novel 5T cell, bit lines of the proposed cell are precharged to ground.

   1. Standby Mode

      Before discussing the operation of proposed SRAM cell, operations of the previously introduced novel 5T cell will be reviewed to clarify the difference between former and later. In the novel 5T cell, introduced earlier [4], when the cell is in a stand by cycle, M5 is turned off by keeping word line (WL) at ground, the bit lines are precharged to VDD, and the data is preserved by the cross-coupled inverters. In the proposed 5T cell, as shown in Fig. 3, during stand by period (precharge stage) the word line (WL) associated with M5 is set to low, which turns off M5, and bit lines are precharged to ground so that the data which was written during write operation is retained by the cross-coupled inverters.

   2. Write Operation

      The write operation is accomplished by effectively asserting the word line (WL). Simultaneously, depending on

      Fig. 3. Proposed 5T SRAM Cell

      the state already stored in the cell, either write 0 or write 1 signal is activated to push one of the bit lines to approximately 2/3 VDD so that the contents of the cell will flip to reflect the bit line data. Consider the situations for the two possible write operations that can be performed on the cell:

      Write 0 Operation

      Assume that initially, i.e. before write 0 operation, the values of the Q and Q_b of the cell are at 1 and 0 respectively. In this stage, transistors M2 and M3 are in the triode region, and M1 and M4 are in cut-off. The operation of write 0 is accomplished by forcing BL_b to approximately 2/3VDD by turning on both the PMOS transistors associated with write 0 and EN signals. Now the source voltage of the NMOS transistor M3 is at approximately 2/3 VDD rather than 0, and there is a charge transfer between input terminals of the inverters because of turn on transistor M5. Thus Q_b is getting charged towards VDD due to M3, which is conducting in the triode region. When the voltage at Q_b exceeds the threshold voltage of M1, the voltage at Q starts discharging towards 0. This initiates a regenerative effect between the two inverters [5]. Eventually, M2 turns off and the voltage at Q falls to 0 due to the pull-down action of M1. Simultaneously, M4 turns on and the voltage at Q_b rises to VDD due to the pull- up action of M4. When the cell finally flips to the new state, the word line associated with M5 is returned to its low stand by level.

      Write 1 Operation

      Assume that initially, i.e. before write 1 operation, the values of the Q and Q_b of the cell are at 0 and 1 respectively. In this stage, transistors M1 and M4are in the triode region, and M2 and M3 are in cut-off. The operation of write 1 is accomplished by forcing BL to approximately 2/3 VDD by turning on both the PMOS transistors associated with write 1 and EN signals. Now the source voltage of the NMOS transistor M1 is at approximately 2/3 VDD rather than 0, and there is a charge transfer between input terminals of the inverters because of turn on transistor M5. Thus Q is getting charged towards VDD due to M1, which is conducting in the triode region. When the voltage at Q exceeds the threshold voltage of M3, the voltage at Q_b starts discharging towards 0. This initiates a regenerative effect between the two inverters [5]. Eventually, M4 turns off and the voltage at Q_b falls to 0 due to the pull-down action of M3. Simultaneously, M2 turns on and the voltage at Q rises to VDD due to the pull-up action of M2. Both write operations are clearly shown in Fig. 4.

   3. Read Operation

   The read operation is achieved simply by asserting the word line (WL), which is associated with the additional transistor M5. Consider the situations for the two possible read operations that can be performed on the cell:

   Read 0 Operation

   Assume that a 0 is stored in the cell, which implies that Q and Q_b of the cell are at 0 and 1 respectively. Therefore, transistors M1 and M4 are in the triode region and M2 and M3 are in cutoff. Initially, BL and BL_b are precharged to a low voltage around ground by a pair of column pull-down transistors as shown in Fig. 3. The word line (WL), held low in the stand by state, is now raised to VDD which turns on additional transistor M5, which in turn creates a current path from the bit line to VDD through the cell. This results in the voltage at Q increases to a little amount from ground and at the same time the voltage at Q_b decreases by a little amount from VDD. The voltage at Q is transferred immediately to BL due to M1, which is conducting in the triode region. Meanwhile, on the other side of the cell, the voltage on BL_b remains low since the column pull down transistor dominates the transistor M3, which is conducting at the edge of the cut-off region. The difference between BL and BL_b is fed to a sense amplifier in a proper manner to generate a valid low output, which is then stored in a data output buffer. In contrast to conventional 6T SRAM cell, here the bit lines are fed to the sense amplifier in inverted fashion to read proper data from the cell. Upon successful completion of read cycle, the word line (WL) is returned to its low stand by level and both the bit lines are precharged back to a value around ground.

   Read 1 Operation

   Assume that a 1 is stored in the cell, which implies that Q and Q_b of the cell are at 1 and 0 respectively. Therefore, transistors M2 and M3 are in the triode region and M1 and M4 are in cutoff. Initially, BL and BL_b are precharged to a low voltage around ground by a pair of column pull-down transistors as shown in Fig. 3. The word

   Fig. 4. Waveforms for write and read operations of proposed 5T SRAM

   line (WL), held low in the stand by state, is raised to VDD which turns on additional transistor M5, which in turn creates a current path from the bit line to VDD through the cell. This results in the voltage at Q_b increases to little amount from ground, and at the same time the voltage at Q decreases by a little amount from VDD. The voltage at Q_b is transferred immediately to BL_b due to transistor M3, which is conducting in the triode region. Meanwhile, on the other side of the cell, the voltage on BL remains low since the column pull down transistor dominates the transistor M1, which is conducting at the edge of the cut-off region. As mentioned in read 0 operation, here also the difference between BL and BL_b is fed to a sense amplifier in a proper manner to generate a valid high output. The length of the additional transistor M5 is 3-4 times longer than the minimum length (Lmin), and all the transistor sizes have been properly designed so that the cell can preserve data during read operation. Both the read operations are clearly shown in Fig. 4.

3. STATIC NOISE MARGIN ANALYSIS OF PROPOSED FIVE TRANSISTOR SRAM CELL

   The stability and robustness of a proposed SRAM cell is usually evaluated by analyzing both its dynamic and static behavior during the write, read, and hold operations. Stability of the memory cell can be estimated from the static noise margin (SNM) analysis. SNM is defined as the minimum DC noise voltage needed to flip the cell state [6], and is used to quantify the stability of the SRAM cell using a static approach. A significant effort has been devoted to explore the impact of process variations using the SNM. Here about proposed 5T SRAM cells static stability during read and hold period has been presented, and the differences between SNM during hold and read modes are compared. The read mode is usually identified as the cells weakest mode.

   1. SNM During Hold Mode

      The SRAM cell immunity to static noise is measured in

      terms of SNM that quantifies the maximum amount of voltage noise that the cell can tolerate at the output nodes of the cross-coupled inverters without flipping the cell. The graphical method to determine the SNM uses the static voltage transfer characteristics (VTC) of the SRAM cell inverters.

      Fig. 5 superimposes the VTC of one inverter to the inverse VTC of the other inverter. The resulting two lobed graph is called a butterfly curve and is used to determine SNM. Its value is defined as the side length of the largest square that can be fitted inside the lobes of the butterfly curve [6]. Fig. 5 shows that the variation of the butterfly curves for two supply voltages (VDD, 2/3 VDD) during hold operation. It clearly shows that lowering the power supply voltage reduces the SNM.

   2. SNM During Read Mode

      Static noise margin is a key performance factor to estimate the ability of the cell that can preserve data during the read operation. SNM during read can be evaluated from voltage transfer characteristic curves obtained by setting word line (WL) to high, while both the bit lines are precharged to a low voltage around ground. Generally, SNM during read takes its lowest value and the cell is in its

      Fig. 6. Butterfly curves during Read operation

      cell with cell ratio (r) of 2 are tabulated in Table I. Cell ratio(r) is defined below:

      weakest state.

      The butterfly curves, shown in Fig. 6, are formed by superimposing of both inverter VTCs taken under read

      =

      (1)

      operation. Fig. 6 shows the variation of SNM for two power supply voltages during read operation, and degradation of the static noise margin with reduction of power supply voltage.

      The SNM during hold and read operation for two different power supply voltages corresponding to an SRAM

      Fig. 5. Butterfly curves during Hold operation

      Where, driver is the transconductance of the storage transistors and access is the transconductance of the access transistors. M1 and M3 act as access transistors and M2 and M4 act as storage or driver transistors in the proposed design. The SNM reduction during read operations with respect to hold operations is considerable at each supply voltage.

   3. Impact of Power Supply Voltage Modulation on Read SNM

      Fig. 7 shows that the impact of power supply voltage reduction on read SNM under various process corners. It is clear that irrespective of the process variations, SNM is reduced significantly with the reduction of power supply voltage. Hence, it is more preferable to maintain full VDD while reading the memory.

4. SIMULATION RESULTS

   The above implemented proposed 5T SRAM with cell ratio (r) of 2 was simulated along with conventional 6T and novel 5T SRAM cells in 180 nm CMOS process using Cadence Spectre and BSIM3v3 models.

   TABLE I. SNM DURING HOLD AND READ OPERATIONS

   VERSUS POWER SUPPLY VLTAGE

   Mode of operation	SNM@VDD (mV)	SNM@2/3 VDD ((mV)
   HOLD	600	480
   READ	432	348

   Fig. 7. Impact of Power Supply Voltage Reduction on Read SNM

   The layout of the proposed 5T SRAM cell is shown in Fig.8, and all the parasitic capacitances, which were extracted from the layouts, along with some additional bit line capacitance approximately 100 fF are included during simulations. The comparison of the proposed 5T SRAM to the conventional 6T and novel 5T SRAMs are tabulated in Table II and Table III respectively. It is observed that the proposed 5T SRAM cell shows good stability over both the conventional 6T and novel 5T SRAMs with better performance and low power consumption.

   TABLE III. PERFORMANCE COMPARISON BETWEEN PROPOSED 5T

   AND NOVEL 5T SRAM CELLS

   Metric	Novel 5T SRAM	Proposed 5T SRAM	% Improvement
   Read SNM	400 mV	432 mV	7.4
   Write Delay	300 ps	101.41 ps	66.19
   Read Delay	469 ps	303.46 ps	35.29
   Power Consumption	122.94 µW	117.1 µW	4.75

5. CONCLUSION

With the aim of attaining a high stability and better performance SRAM, a five transistor (5T) SRAM cell is designed and simulated using a 180 nm CMOS process. The proposed cell exhibits 22.66% better performance with respect to conventional 6T, and 35.29% improvement with respect to novel 5T SRAM cell. Read static noise margin of the proposed cell is 40.97% higher than the conventional 6T SRAM cell with significant reduction in power consumption. Simulated results, as seen from process corner analysis, quite well justify the robustness of the design even at worst case process variations. Therefore, the proposed 5T SRAM cell design would be suitable for various high speed and low power embedded cache, stand-alone IC, and network applications.

Fig. 8. Layout of the Proposed 5T SRAM Cell

TABLE II. PERFORMANCE COMPARISON BETWEEN PROPOSED 5T

AND CONVENTIONAL 6T SRAM CELLS

ACKNOWLEDGMENT

With warm Thanks to RTTC BSNL Staff, Mysore and My Family members & Friends for supporting me.

References

1. A. Agarwal, C. H. Kim, S. Mukhopadhyay, and K. Roy, Leakage in Nano-Scale Technologies: Mechanisms, Impact and Design Considerations, Proc. of the 41st Design Automation Conference (DAC04), June 2004, pp. 6-11.

2. S. Mukhopadhyay, H. Mahmoodi, and K. Roy, Modeling of Failure Probability and Statistical Design of SRAM Array for Yield Enhancement in Nanoscaled CMOS, IEEE Trans. on Computer- Aided Design of Integrated Circuits and Systems, Vol. 24, no. 12, p.p 1859-1880, December 2005.

3. Debasis Mukherjee, Hemanta Kr. Mondal, and B.V.R. Reddy, Static Noise Margin Analysis of SRAM Cell for High Speed Application, IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 5, September 2010.

4. Michael Wieckowski, Martin Margala, A novel five-transistor (5T) SRAM cell for high performance cache, Proc. of the IEEE International SOC Conference, Dec 2005, pp.101-102.

5. David A. Hodges, Horace G. Jackson, Resve A. Saleh, Analysis and design of digital integrated circuits: In Deep Submicron Technology, McGraw-Hill Edition, 2004.

AUTHORS:

Dachineni Bharath Satyanarayana & Kakarlapudi Pradeep are students at K L University.

Tsr Prasad & T Ravi are Asst. professor at K L Universty.

Phone no:+917204199074