ALPHACORE INC — Department of Defense SBIR Phase I: HR0011SB20224-15

Description

Heterogeneous Integrated Circuits (HIC) combine different ICs into a single package. In this DARPA Phase I SBIR, Alphacore and our partner Riscure will conduct a feasibility study to create a method of identifying which ICs are in a package through available side channels, without opening the package. A Grid-Array Side-channel Probe (GASP) system will be studied for possible development, comprising as many as 16 coil, with each coil simultaneously capturing the electromagnetic (EM) signals from the HIC. The HIC is provided with a stimulus (e.g., power on, or starting a particular device function), and the resulting electromagnetic side channels are captured. We are developing this coil matrix specifically for this research, while the capture and analysis equipment will be based on Riscure’s Inspector SCA tool suite.  In Phase I, the team will explore six aspects of investigation and classification of side channels in a HIC:  Validation vehicle creation  Signal capture (pre- or post-silicon)  Source separation  Feature extraction  Learning  Classification  Pre-silicon capture of signals can be used during the learning phase if the design details of the target are available, e.g., in the malicious behavior monitor use case. Pre-silicon signal capture is done by leveraging the existing simulation techniques of the DARPA SCATE program to capture traces for individual ICs.  As the number of templates grows, neural networks can be leveraged both in the feature extraction phase and the learning and matching phase, allowing generalization from ICs that have been characterized, to unknown ICs.  The classification takes as input an ‘unknown’ feature vector and the learned templates. It will provide a confidence score for each template, and thereby which IC / function is most likely. For the identification use case, the output is which IC / function is detected for each of the 16 signals that will be probed individually by the GASP system. For the malicious behavior monitor use case, the output is a confidence value for each of the 16 of observing malicious behavior.   As described, this method is a framework that can be extended with more templates, improved feature detection, different side channel sources, and so on, hence realizing a future (iterative) improvement path. We expect the method above to be able to classify with great accuracy both which ICs are in a HIC, as well as malicious behavior of a HIC.?  DARPA’s stated objectives for the Phase I study are to demonstrate a framework with an 80% probability of detecting various functions and components, and a 0.01% false positives rate, on a HIC with 2 ICs. The framework should be extendable, e.g., with improved analysis techniques and different side channel sources. Based on our plan, we believe that these objectives will be met. (For a follow-up Phase II, a 90% detection rate, 0.01% false positive rate on a 4 IC HIC is targeted.)