Abstract
This paper proposes an improved technique for implementing MUX-FSM-based stochastic computing (SC) in on-device neural networks. The proposed method focuses on enhancing the speed of SC operations, which is crucial for real-time applications on resource-constrained devices. By optimizing the MUX-FSM design, the authors aim to reduce latency and improve the overall efficiency of SC-based neural network accelerators.
Keywords
Stochastic Computing (SC), MUX-FSM, On-device Neural Networks, Hardware Acceleration, Low-power AI
Full Text
This paper proposes an improved technique for implementing MUX-FSM-based stochastic computing (SC) in on-device neural networks. The proposed method focuses on enhancing the speed of SC operations, which is crucial for real-time applications on resource-constrained devices. By optimizing the MUX-FSM design, the authors aim to reduce latency and improve the overall efficiency of SC-based neural network accelerators.