Enhancement in IoT through Custom Instruction Set Architectures and TinyML: Review

The Internet of Things or IoT means to the confluence of the network of interconnected devices, the technologies that enable communication among these devices, and between these devices and the cloud. IoT contains embedded CPUs for performing a variety of functions. The processors for enhancing IoT applications form an active area of research in the domain of Computer Architecture. Embedded processors are resource and power-constrained challenging the execution of deep learning and machine learning algorithms. One proposed solution in literature is TinyML (Tiny Machine Learning) to overcome this limitation. TinyML architecture for IoT aims to deliver low latency, and effective bandwidth usage, reinforce data security, enhance privacy, and save costs. TinyML enables IoT devices to perform ML computations to be performed in the edge and avoid frequent access to the cloud which increases the performance. This paper presents a review of processor designs that are designed for TinyML architecture. For addressing other important issues in IoT a review of custom ISAs (Instruction Set Architectures) developed is also presented in this review.