GLASS-CV: Galois Low-energy Asynchronous Secure SoC for Computer Vision
Using a novel, cloud-based secure microelectronics design environment developed by the Air Force Research Laboratory for the Microelectronics Design and Prototyping Challenge and funded by the OSD Trusted and Assured Microelectronics Program, GLASS-CV system-on-a-chip (SoC) aims to produce a low-power, secure processor specifically targeted for autonomous systems where power consumption is a challenge.
Information gathering networks such as sensor networks, security systems, and more recently autonomous systems can collect substantial amounts of information, most of which are of limited value and prohibitively costly to transfer back to a data center. Ideally, data processing needs to happen at the network edge; only when an event of interest is detected should a message be propagated to the data center.
However, processors in edge nodes often have limitations that make efficient data analysis impossible: they might have power constraints, require special thermal considerations, etc.
On the GLASS-CV project, we aim to create a system that can provide basic artificial intelligence (AI) and machine vision (MV) capabilities to power and thermally constrained systems. In other words, we aim to bring some of the capabilities of a mobile phone processor within a power envelope that is closer to that of a present-day microcontroller.
The system incorporates:
- a neural network accelerator for efficient AI/inference operations,
- a RISC-V general-purpose processor,
- a secure JPEG core for image processing, and
- a formally verified cryptographic accelerator providing data protection.
The SoC also features a secure boot subsystem for anti-tamper protection, which can detect if the processor is compromised.
https://www.youtube.com/watch?v=F_HWde73VeQ
Challenges and innovations
There are several processors in the market that are capable of handling workloads for autonomous applications. However, most of these solutions rely on general-purpose architectures that cannot efficiently execute such tasks. Our processing system is architected with such applications in mind and thus comprises a rich set of co-processors that are targeted for these kinds of workloads and each coprocessor is optimized for their individual task.
Additionally, our system relies on a pair of core technologies from our team’s expertise to bring unique capabilities to the end system using it:
- We use resilient asynchronous logic to design our processor, which enables operation at very low voltages (as low as near-threshold). This allows us to achieve unparalleled energy efficiency and a wider operating range in terms of environmental conditions.
- We aim to leverage our security expertise to harden this processor with a variety of security features so that it can communicate securely and also detect tamper events and resist data security breaches.