Integrated Signal Processing Systems for the Internet-of-Things

点击翻译按钮获取中文摘要

The Internet-of-Things (IoT) is the interconnection on a massive scale of autonomous smart objects, enabling innovative applications in health, infrastructure, environment, agriculture, and industry. The goal of the proposed research program is to develop signal processing hardware for smart objects in the IoT addressing two critical challenges: energy efficiency and security. We will investigate fundamentally new approaches addressing the very tight cost, size, and power budgets of smart objects in the IoT that cannot be met by conventional architectures and methodologies by developing neuromorphic ("brain-like") low-power signal processing hardware. The proposed methodology is organized around three research thrusts: i) stochastic computing for low-power signal processing in smart objects, ii) associative memory for secure network processors in the IoT, and iii) polar codes for secure, low-complexity wireless IoT communications. Stochastic computing is the design of hardware that computes with random pulse trains, resulting in remarkably simple logic circuits that offer a natural way to trade-off power consumption with the precision of a computation. Another brain-inspired structure, associative memory, integrates memory with searching and maps naturally to very efficient logic-in-memory architectures. Both of these techniques are inherently fault-tolerant and therefore able to gracefully accommodate errors induced when reducing the supply voltage to minimize power. Polar codes are a breakthrough capacity-achieving error-correction technique for digital communications that can provide secure wireless communications, with a low-complexity decoding algorithm. This work has the potential to introduce a disruptive paradigm for low-power signal processing systems design. In the short term, our approach has the potential to greatly reduce power consumption of critical signal processing components for smart objects in the IoT, demonstrate the advantages of polar codes, driving them towards standards, and train highly qualified personnel to be employed in Canadian industry. In the long term our work will help demonstrate the potential of neuromorphic architectures for emerging computational problems.