混合机器学习和数值优化方法用于模拟电路去混淆.pdf

1、A Hybrid Machine Learning and Numeric Optimization Approachto Analog Circuit DeobfuscationDipali Jain,Guangwei Zhao,Rajesh Datta,Kaveh Shamsi22.01.2025Department of Electrical And Computer EngineeringUniversity of Texas At Dallas,Richardson,USA Background and Motivation Proposed Method Experiments a

2、nd results Discussion ConclusionContentsASP-DAC 20252Oracle Guided Circuit Learning(OGCL)ae(k,x)ae(k*,x)x(t)y(t)x(t)y(t)kk*where k*=unknown analog valuesx=controllable analog inputsae=analog circuity=output responseObfuscated(Analog)CircuitOracle(Analog)Circuitae(k,x)ae(k,x)ASP-DAC 20253Applications

3、 of OGCLSecurity Analysis of circuit obfuscationPredicting Unknowns that are obscured in reverse engineering1Trojan Detection 21 Jain,Dipali,Guangwei Zhao,Rajesh Datta,and Kaveh Shamsi.Towards Machine-Learning-based Oracle-Guided Analog Circuit Deobfuscation.In 2024 IEEE International Test Conferenc

4、e(ITC),pp.323-332.IEEE,2024.2 Datta,Rajesh Kumar,Guangwei Zhao,Dipali Jain,and Kaveh Shamsi.On Hardware Trojan Detection using Oracle-Guided Circuit Learning.In Proceedings of the Great Lakes Symposium on VLSI 2024,pp.198-203.2024.ASP-DAC 20254Analog Circuit Obfuscation.Obfuscating a current mirror

5、circuit used to bias an amplifierASP-DAC 20255Equation(Analytical)Based ApproachStatic:Nonlinear Programming(NLP)Time-independent analog circuit:ae(k*,x)(xq,yq)x1(t)x2(t)x3(t).xn(t)y1(t)y2(t)y3(t).yn(t)kfj(xn,kn,sn)=yn xn=query inputs kn=unknowns sn=intermediate variables yn=observablesTime-dependen

6、t analog circuit:Dynamic:Differential-Algebraic Equations(DAEs),with time derivatives of x(t),y(t),s(t)ASP-DAC 20256Optimal Control ApproachFinding the optimal value of the setof unknown parameters k,Solutions to DAEs are sufficientlyclose to a set of desired curves.Larger systems involve performing