A bit-serial computing accelerator for solving 2D/3D laplace and poisson equations

Abstract

Partial differential equations (PDEs) play a pivotal role in various scientific and engineering fields due to their capability to model and solve numerous real-world problems. For edge devices, the need for low-power consumption and real-time processing is crucial. This thesis proposes a bit-serial computing accelerator, accompanied by simulation results, that demonstrates a compact chip area, low power consumption, and high computational efficiency. Furthermore, several optimization techniques have been integrated to enhance performance. Potential future research directions are also discussed, including memory compression strategies and the integration of physics-informed neural networks (PINNs).