MIT Tang Family FinTech Undergraduate Research and Innovation Scholar
Lightweight Cryptography for Energy-Efficient and Secure Deep Neural Network Accelerators
Anantha P. Chandrakasan
Deep neural networks (DNNs) are increasingly deployed in security-critical embedded applications, such as autonomous driving and biometric authentication. However, security and privacy of those applications can be undermined by several attacks targeting hardware-level vulnerabilities. This motivates the need to encrypt the values stored in the chip, so that only the person with the right key will be able to access it, thus protecting the data stored in the memory of the chip (say parameters of the DNN like the weights and biases) from outside attackers. Lightweight Cryptographic (LWC) methods are of special interest because of low power consumption, and this project aims to implement and compare such methods to ensure that data remains secure within the chip in an efficient manner.
I am participating in this SuperUROP because I have always been fascinated by computer architecture, machine learning and cryptography. While I had experience in these fields independently through classes or research, this is an opportunity to combine all of these seemingly interdisciplinary fields in an effort to develop energy efficient accelerators. I am excited to learn more about ML accelerators and lightweight cryptography!