ENERGY-EFFICIENTAPPROXIMATE CIRCUITS THAT CAN BE ADJUSTED FOR SELF-POWERED AI AND AUTONOMOUS EDGE COMPUTING SYSTEMS

Abstract

The growing reliance of edge artificial intelligence (AI) systems on self-powered or energy-constrained hardware platforms necessitates extremely efficient processing units without sacrificing functional precision. Optimizing multiply-and-accumulate (MAC) operations is crucial for sustainable edge intelligence since they account for the majority of energy consumption in contemporary AI workloads. In order to enable dynamically adjustable accuracy–energy trade-offs for MAC architectures, this work presents unique tunable approximation compressor designs. Low-complexity approximation techniques and tenability knobs that enable real-time configuration based on workload precision needs or available energy budgets are features of the suggested compressors. The designs maintain sufficient accuracy for errorresilient AI workloads while achieving significant savings in power consumption, latency ,and silicon area when incorporated into MAC units. The designs' effectiveness and viability are validated by hardware implementation and assessment using Xilinx Vivado, indicating their appropriateness for energy-adaptive computation in next- generation edge AI and autonomous sensing platforms.