@techreport{oai:ipsj.ixsq.nii.ac.jp:00210528, author = {井阪, 友哉 and 新谷, 道広 and アフメド, フォイサル and 井上, 美智子 and Yuya, Isaka and Michihiro, Shintani and Foisal, Ahmed and Michiko, Inoue}, issue = {30}, month = {Mar}, note = {Field-programmable gate-array(FPGA)の回路性能は使用により経時的に劣化することから,FPGA 上に設計したリング発振器(Ring oschillator,RO)の周波数を解析し,再利用か否かを機械学習により判定する手法が提案されている.ところが,既存の再利用 FPGA 検出手法の多くは,大量の新品 FPGA を機械学習で使用する正解データとして準備する必要があり,実際の開発現場への適用に課題がある.そこで,本稿では,正解データを必要としない教師なし再利用 FPGA 検出手法を提案する.提案手法は,FPGA の全論理ブロックに多段 RO を設計し隣接RO を比較することで製造プロセスばらつきによる影響を低減して経年劣化による影響を顕在化した上で,直接密度比推定による異常値選別を適用する.同型の市販 FPGA10 個のうち 2 個を実動作により劣化させた評価実験では,提案手法はこの 2 個の再利用 FPGA を誤分類なしに検出することができた., It is well known that the performance of field-programmable gate-array (FPGA) degrades over time due to their usage. Several methods have been proposed in which the frequency of a ring oscillator (Ring oscillator, RO) designed on an FPGA is analyzed and whether it is recycled or not is determined by machine learning model. However, most of the existing methods require a large amount of fresh FPGAs as correct data for machine learning, which poses a problem in actual application to industrial. In this paper, we propose a novel unsupervised recycled FPGA detection method. In the proposed method, multi-stage ROs in all logical blocks are designed on the FPGA, and adjacent ROs are compared to reduce the effects of manufacturing process variation and highlight the effects of aging degradation in the measured frequencies, and then apply outlier detection through direct density ratio estimation. Through experiments using 10 fresh commercial FPGAs and 2 out of them that were aged, we demonstrate the proposed method can detect these two recycled FPGAs without misclassification.}, title = {自己参照に基づく直接密度比推定を用いた教師なし再利用FPGA検出}, year = {2021} }