Spintronic Solutions for Approximate Computing

In the conventional computing systems, high redundancy in terms of energy consumption, speed and area are usually applied to guarantee accurate operations. However, energy consumption has become a bottleneck for further scaling down of silicon-based devices, which can no longer be increased to ensure accuracy. In order to find a balance between accuracy and other performance metrics, approximate computing has been intensively studied and applied in the applications that are intrinsically fault tolerant. Featured with non-volatility, fast access speed, high scalability and current-induced thresholding operation, spintronic devices are promising candidates for approximate computing. This chapter exploits the application of approximate techniques in the spintronic device-based circuit designs for energy-efficient processing in-memory. Approximate techniques based on spintronic devices are explored for both traditional full adders and write-only bitwise full adders with reduced complexity. The simulation results show that the energy can be significantly reduced with negligible loss in output quality.