Reduced power consumption compute-in-memory system, method of operating same
AssigneeTaiwan Semiconductor Manufacturing Company, Ltd.
InventorsHidehiro Fujiwara, Je-Min Hung, Haruki Mori
Life of this patent
Estimating the term…
Loading patent…
AssigneeTaiwan Semiconductor Manufacturing Company, Ltd.
InventorsHidehiro Fujiwara, Je-Min Hung, Haruki Mori
Estimating the term…
A compute-in-memory (CIM) system includes a first leader-follower (L/F) flip-flop (FF), a suspender unit and a multiplication unit. The first L/F FF is configured to receive a data signal. The first L/F FF includes: a first leader FF configured to receive the data signal and a first follower FF configured to generate a signal first_Q_follow that represents a output signal of the first L/F FF. The multiplication unit is configured to receive a weight signal W and the signal first_Q_follow and generate a first product signal. The suspender unit is configured, during a given cycle of the first clock signal, detect that a first scenario is true in which all bits b(k) of the data signal equal logical zero and accordingly suppress latching from being performed by the first L/F FF, and control the multiplication unit to generate the first product signal as being equal to logical zero.
A compute-in-memory (CIM) system comprising:
A method of reducing power consumption in a compute-in-memory (CIM) system, the method comprising:
A compute-in-memory (CIM) system comprising:
Dependencies are inferred from the claim text.