Neural Network Accelerator
Neural network accelerators are specialized hardware designed to significantly speed up and reduce the energy consumption of deep learning computations, moving processing from cloud servers to resource-constrained edge devices. Current research emphasizes efficient memory hierarchies, novel arithmetic techniques (like digit-serial processing and power-of-two quantization), and algorithm-hardware co-design to optimize performance for various model architectures, including convolutional neural networks, transformers, and spiking neural networks. These advancements are crucial for enabling real-time AI applications in areas like embedded systems, mobile devices, and safety-critical systems, where low power and high speed are paramount.
Papers
NEON: Enabling Efficient Support for Nonlinear Operations in Resistive RAM-based Neural Network Accelerators
Aditya Manglik, Minesh Patel, Haiyu Mao, Behzad Salami, Jisung Park, Lois Orosa, Onur Mutlu
PhotoFourier: A Photonic Joint Transform Correlator-Based Neural Network Accelerator
Shurui Li, Hangbo Yang, Chee Wei Wong, Volker J. Sorger, Puneet Gupta