Method for obfuscation of hardware
The method for obfuscating hardware partially imitates the neural network perceptron, obfuscating the hardware design. This method obfuscates the design functionality and immunes integrated circuits against Trojan insertion. This method can also be used to check for the existence of faults inside chips. This method resolves the concern related to security and reliability when outsourcing the manufacture of integrated circuits.
1. A method for obfuscating a hardware of an integrated circuit comprising:
wherein the integrated circuit comprises:
two or more logic gates;
two or more transition nets;
a training pattern generator;
a control unit; and
at least one trainable logic block, comprising:
at least one input; and
a configuration line;
inserting at least one isolation point into at least one output of the trainable logic block;
identifying at least one candidate logic gate;
identifying at least one candidate interconnect;
using an output of the at least one candidate logic gate and an output of the at least one candidate interconnect as inputs for the trainable logic block;
tangling the at least one candidate logic gate with at least one other logic gate;
mixing at least one low transition net of the candidate interconnect with at least one other low transition net, wherein any remaining low transition nets comprise inputs to the trainable logic block, wherein a low transition net comprises a net that has a low transition activity;
providing nonvolatile memory, wherein said nonvolatile memory protects at least one configuration line; and
training the at least one trainable logic block, comprising:
providing three modules comprising:
the training pattern generator;
a response analyzer; and
a weight update unit;
selecting a training pattern;
selecting a corresponding value for the training pattern;
generating of an input pattern by the training pattern generator;
isolating the at least one trainable logic block to be trained; and
applying the training pattern individually to each at least one trainable logic block.
2. The method of claim 1 , wherein the at least one isolation point is inserted at at least one output of the trainable logic block.
3. The method of claim 1 , wherein the at least one isolation point comprises a pass transistor.
4. The method of claim 1 , wherein the at least one isolation point comprises a high fan-out.
5. The method of claim 1 , wherein the nonvolatile memory comprises floating gate transistors.
6. The method of claim 1 , wherein the training step comprises:
providing three modules comprising:
the training pattern generator;
a response analyzer; and
a weight update unit;
selecting a training pattern;
selecting a corresponding value for the training pattern;
generating of an input pattern by the training pattern generator; and
applying the training pattern to at least one trainable logic block.
7. The method of claim 1 , wherein the training step comprises:
providing three modules comprising:
the training pattern generator;
a response analyzer; and
a weight update unit;
selecting a training pattern;
selecting a corresponding value for the training pattern;
generating of an input pattern by the training pattern generator; and
applying the training pattern to at least one trainable logic block by the training pattern generator.