Skanda Koppula
MIT EECS — Analog Devices Undergraduate Research and Innovation Scholar
Secure Voice Recognition Chip based on Homomorphic Encryption
2015–2016
Anantha P. Chandrakasan
Homomorphic encryption is a method of encryption that allows computing over ciphertext; not surprisingly, this form of encryption could be highly applicable in situations where untrusted parties provide important computing services on personal (or otherwise sensitive) data. Unfortunately, high-level implementations of homomorphic cryptosystems are slow and often power intensive. A homomorphic system in an application specific circuit, on the other hand, may allow us to increase data throughput while also using less power. Specifically, this project aims to prototype an ASIC-based voice recognition service that authenticates users based on a voice sample. Our system will compare a protected voice model with input voice samples
What makes me most excited is the real applicability of the project, and the challenging work I know is up ahead. I owe a lot of my background preparation to many of the classes I’ve taken here; thank you, 6.875 and 6.175. Ultimately, I hope to learn about (1) complexities in homomorphic cryptosystems and (2) more on designing great chip architecture