W5170124C0128

Small Business Innovation Research Program (SBIR) Phase II

Synthetic Data Generation for Enforte Attest

With ongoing microelectronic supply chain issues, the demand for genuine field-programmable gate arrays (FPGAs) is increasing but so is the occurrence of counterfeit devices. Frequently, devices are used, salvaged from old systems, and repackaged as new. These recycled devices represent the largest class of counterfeit devices and are becoming more rampant. Therefore, it is often necessary to test whether a device is counterfeit before employing it in a new system. Current methods for evaluating the genuine nature of devices are frequently destructive, allowing for only small sample testing within lots. Other methods require complex external equipment and cannot be readily deployed throughout the supply chain. Graf Research Corporation has developed a methodology for using telemetry sensors to characterize an FPGA device and subsequently classify whether a device is a repackaged counterfeit via statistical and machine learning models. The new method utilizes minimal external equipment, is non-destructive, and can be employed at any point throughout the supply chain. Artificial Intelligence (AI) and Machine Learning (ML) approaches to rapid decision making are becoming increasingly prevalent in the area of microelectronics validation. The ability to rapidly determine a chip is not counterfeit or modified is a key aspect of ensuring systems will faithfully serve the warfighter. Producing AI solutions to microelectronics evaluation requires a significant dataset for training, however, and the current process of manually gathering diverse datasets made up of physical chips to produce a representative and sufficiently diverse dataset is prohibitively expensive in time and material costs. It also becomes increasingly difficult to procure known good samples as analyses turn to older devices with limited supply. The limitations on developing AI methods imposed by a lack of data, are overfitting of models that do not perform well on unseen test or real-world data. This is due to ML models tendencies to memorize smaller datasets. With this in mind, the need for synthetic data to assist in training is becoming increasingly important as a part of building AI solutions to perform identification and classification of user configurable microelectronics for the purposes of rapid microelectronics assurance or counter fit detection. Solving this need to support AI solutions can reduce the manual effort and expense required to determine if a part is genuine or how much life the part has left. The Phase II effort proposed here integrates novel synthetic data generation for FPGA sensors with Graf Research s Enforte Attest counterfeit detection platform. The Phase II program advances Phase synthetic data generation capabilities, improves the efficacy of generated data, integrates the data generation capabilities into the counterfeit detection platform and demonstrates prototype functionality on a modern FPGA.

The goal of phase II is to continue the R&D efforts initiated in Phase I. Funding is based on the results achieved in Phase I and the scientific and technical merit and commercial potential of the project proposed in Phase II.

A234-007

A234-007-0287

23.4

Blaise Zandoli