Finding hardware Trojans in chips with ML
Amida Technology Solutions, a DC open-source data interoperability and security startup backed by In-Q-Tel, Blue Delta Capital, Red Hat and DoD
Results
- Prototype delivered
- Phase I
- Trojan and trigger types modeled
- 10
- Patent applications (2018 to 2020), Amida hardware team
- 3
- HST 2018 paper on Trojan detection through emulation
- IEEE
The Phase I prototype worked, and a Phase II prototype was in development by mid-2020. The Amida hardware team filed 3 patent applications and published at IEEE HST 2018.
The problem
Chips built through outside supply chains can carry hidden circuits or counterfeits that standard testing misses. For naval systems, medical devices, autonomous vehicles and industrial controls, one compromised chip puts the whole system at risk.
The build
Phase I emulated a clean "golden model" of a system on an FPGA, inserted ten digital, analog and environmental Trojans and triggers, and collected behavior data through custom on-chip instruments. The proof of concept covered data leakers, behavior modifiers and reliability attacks, plus side-channel Trojans on temperature, voltage, IR and analog waveform. ML models trained on clean and compromised data detect, locate and size the attack and return a Trojan or counterfeit signature. Phase II, the Trojan Emulation and Exploration Environment (TE3), adds scale, automation, remediation and a library of models that predict anomalies.
Key calls
Framed the product around supply-chain risk.
Trojans can enter anywhere from design to packaging, and COTS parts in defense systems carry that risk. The pitch led with where Trojans come from.
Owned the line from requirements to demos.
Requirements, proposals, hardware assurance collateral and product demos to partners and clients.