AlPicoSoC: A Low-Power RISC-V Based System on Chip for Edge Devices with a Deep Learning Accelerator

In recent years, the integration of Artificial Intelligence (AI) into Internet of Things (IoT) devices has gained significant importance due to its necessity in serving human-centric applications. However, these devices impose stringent hardware requirements in terms of energy, area, and efficient computing capacity. This paper presents an IoT System-on-Chip (SoC) based on the RISC-V architecture, named AlPicoSoC, equipped with a Deep Neural Network (DNN) accelerator known as Alpha Accelerator. Alpha Accelerator is specifically designed to accelerate inference tasks on Deep Learning (DL) models, providing layer-level computation in each working cycle while ensuring minimal resource utilization and power consumption. AlPicoSoC is implemented and evaluated on an Ultra96-V2 FPGA board. Experimental results obtained using the MNIST dataset demonstrate the system’s high accuracy, achieving up to 97.69%. Significantly, AlPicoSoC system outperforms the original PicoSoC system by a remarkable factor of over 1679.47 times, while the resource utilization and energy consumption of AlPicoSoC only marginal increases, with a ratio of just over 8.26 and 1.21, respectively.