SRAM based In-Memory Computing for Energy-Efficient AI Inference