Low-power static random access memory
A low-power static random access memory (SRAM) is set forth which includes a cache memory function without requiring a special bit cell, and which realizes robust read and write operation without any write assist circuit at 16 nm or below FinFET technology. The SRAM comprises a half-Vdd precharge 6T SRAM cell array for robust operation at low supply voltage at 16 nm or below, and with cacheable dynamic flip-flop based differential amplifier referred to as a main amplifier (MA). Prior art 6T SRAM cell arrays use Vdd or Vdd-Vth precharge schemes, and have separate read and write amplifiers. The SRAM set forth uses one main amplifier only, which is connected to the bit line (BL) through a transmission gate. The main amplifiers functions as a read amplifier, write amplifier, and a cache memory.
1. A static random-access memory comprising:
at least one six-transistor memory cell arranged between a first bitline, a second bitline and a word line;
a bitline precharge circuit for precharging the first bitline and second bitline to a voltage of Vdd/2 prior to the at least one six-transistor memory cell receiving a word line signal;
a main amplifier for receiving signals on data lines din and /din in a first voltage domain via a gate WEi; and
a main amplifier precharge circuit for precharging the main amplifier in response to a signal /PEMA such that the main amplifier amplifies signals in the first voltage domain to a second voltage domain, wherein the main amplifier precharge circuit precharges the global bit line busses GBL and /GBL before the main amplifier receives signals on data lines din and /din and independently of the bitline precharge circuit precharging the first bitline and second bitline, wherein a plurality of six-transistor memory cells along a selected word line are simultaneously read or written, wherein the main amplifier also functions as a cache memory.
2. The static random-access memory of claim 1 , wherein the main amplifier amplifies read data on data line dout from the at least one six-transistor memory cell when accessed, and retains the data until the main amplifier is subsequently accessed.
3. The static random-access memory of claim 1 , wherein the main amplifier retains write data when accessed until the main amplifier is subsequently accessed again.
4. The static random-access memory of claim 3 , for use in a N-bit deep learning processing element (DPE) of an AI system, wherein signals /PEMA, WE, ME, and /ME are decoded by N and a plurality of which are asserted to turn on the main amplifier for write masked operation.
5. The static random-access memory of claim 1 , wherein the main amplifier is controlled by a write signal sequence of main amplifier precharge triggered by signal /PEMA, then writing data into the main amplifier triggered by enable signals ME and /ME, followed by the first bitline and second bitline being connected to the global bit line busses GBL and /GBL causing the main amplifier to operate as a write amplifier.
6. The static random-access memory of claim 1 , wherein enable signals ME and /ME are asserted in either standby or active modes and signal /PEMA is asserted before the main amplifier starts reset, such that the main amplifier operates as a cache amplifier.
7. A The static random-access memory of claim 1 , wherein the cache memory is connected to an 8-bit deep learning processing element (DPE) having 32 columns of main amplifiers, wherein one said main amplifier of four of said 32 columns of main amplifiers is selected.
8. The static random-access memory of claim 7 , wherein data written to the main amplifier is selected either from din or latched data depending on assertion of a write mask enable signal WME.
9. The static random-access memory of claim 8 , wherein one latch is provided every four main amplifiers for latching read data of one of the four main amplifiers when an enable signal is de-asserted.
10. A The static random-access memory of claim 1 , wherein the cache memory is connected to a 4-bit deep learning processing element (DPE) having 16 columns of main amplifiers, wherein one said main amplifier of four of said 16 columns of main amplifiers is selected.
11. The static random-access memory of claim 1 , having a plurality of groups of main amplifiers, wherein the read data on dout from each group of main amplifiers charges up a capacitance C0 in response to an S0 timing signal which is then shorted to a plurality of shared capacitances equal in number to the plurality of said groups of main amplifiers in response to a S1 timing signal, and wherein a shorted voltage across the plurality of shared capacitances is compared with the Vdd/2 voltage.
12. The static random-access memory of claim 1 , having a plurality of groups of main amplifiers, wherein the read data on dout from each group of main amplifiers charges up a capacitance C0 in response to an S0 timing signal which is then shorted to a plurality of shared capacitances equal in number to the plurality of said groups of main amplifiers plus one in response to a S1 timing signal, and wherein a shorted voltage across the plurality of shared capacitances is compared with the Vdd/2 voltage.