Exciting news has just emerged in the world of IoT technology: a groundbreaking patent for the NetHD system was granted in June 2024, following the publication of the primary research paper on May 26, 2024. This work was led by Dr. Mohsen Imani, UC Irvine. Mohsen, a former SRC Research Scholar in the STARnet and JUMP programs, is a rising star, recently awarded both the DARPA 2023 Young Faculty Award and the SRC 2023 Young Faculty Award as well as the Office of Naval Research Young Investigator Program award for 2023. Other authors on this paper include UC Irvine PHD students Prathyush Poduval, Yang Ni, and Zhuowen Zou, along with Notre Dame Professor Kai Ni.

PAPER NetHD: Neurally Inspired Integration of Communication and Learning in Hyperspace

PATENTNetwork-based hyperdimensional system, US Patent Number US12015424B2

This work was made in strong collaboration with SRC member companies IBM, Intel, NXP, and Qualcomm. Of the 12 journal publications and more than 20 conference papers resulting from this work, seven were published jointly with liaisons from member companies.

NetHD is a novel system designed to enhance the efficiency and robustness of data communication and learning in Internet of Things (IoT) devices. It utilizes Hyper-Dimensional Computing (HDC), which is inspired by the way the human brain processes high-dimensional data. High-dimensional data refers to data with a large number of attributes or features, making it complex and rich in information, while noise-tolerant data representation means that the system can effectively handle and process data even in the presence of errors or disturbances. HDC is known for its noise-tolerant data representation, making it suitable for the unreliable networks often found in IoT systems.

This work overcomes the challenges faced by current communication systems, such as high communication costs and lack of robustness to noise, which can be particularly problematic for ultra-low power IoT devices. NetHD addresses these issues by using HDC’s redundant and holographic data representation, which allows for a significant number of bits to be corrupted without losing important information. This feature is leveraged to design an iterative decoding method that can transfer data back to the original space without the need for an error correction mechanism.

Furthermore, NetHD integrates data transmission and learning by enabling direct hyperdimensional learning on transmitted data, eliminating the need for data decoding. This results in a system that is not only more energy-efficient but also faster compared to traditional deep neural network (DNN) approaches.

In summary, NetHD offers a new way of thinking about data communication and learning in IoT devices, promising faster and more energy-efficient operations without compromising on robustness or accuracy. It represents a significant step forward in the development of intelligent and sustainable IoT systems.

View Dr. Mohsen Imani's GRC Artificial Intelligence Hardware project 2988.001, https://app.pillar.science/projects/4971/overview.