Configurable 8T SRAM-based Computing In- Memory Architecture for Enabling Shift Operation and Multibit Dot-Product Engines

The conventional computing system in our laptops and mobile phones uses Von-Neumann architecture. Computing In-Memory (CiM) acts as the bottleneck for modern data-intensive applications like artificial intelligence. We have implemented the vector dot-product and shift operation in this work using 8T SRAM for CiM. In order to maintain the precision of a dot-product operation, the cell must be operated in the voltage ranges where the current is a linear function of the applied voltage. We need to size the transistors in the ratio of 8:4:2:1 so that the respective transconductance also remains in the ratio of 8:4:2:1. The summing current of all RBL of four columns yields the required dot product. The left/right shifting operation is implemented with the help of 8T SRAM-based CiM architecture along with a 3×1 multiplexer and D flip flop.