[0002] Nowadays, in order to avoid the threat of reverse engineering to security chips, the most effective method in security chip design is to use Physically Unclonable Function (PUF) to provide the key required for security chip encryption and decryption. PUF detects Due to the random changes in the physical properties of the materials that make up the circuit devices during the production of integrated circuits, it is impossible for even a chip manufacturer to use the same circuit to reproduce the exact same key, and it is even more impossible for attackers to deduce the key through layout analysis.
It can be seen that the use of PUF to provide keys can naturally prevent external reverse engineering attacks, but since PUF provides the keys required for encryption and decryption of secure chips, its output key value must have absolute process randomness and environmental stability Resilience, that is, the output key is different under different processes, but under different voltage and temperature environments, the output key must be absolutely stable
[0003] In order to improve the environmental stability of PUF, the existing research has proposed a variety of measures, but none of them can completely eliminate the unstable unit of PUF, or the elimination cost is too high
In terms of specific circuit optimization, the optimization of delay units and delay stages can only reduce the bit error rates of Transient Effect Ring Oscillator (TERO) and Arbiter PUFs to 6% and 3.2%, respectively. The optimization of RO-type PUF and comparator-type PUF by temperature compensation can only reduce the bit error rate to 3.5% and 4.6%, respectively. The effect of these optimization methods is not obvious, and the bit error rate is still high. And it is easily affected by the process, and it cannot be compensated at any process angle
[0004] In terms of new non-clonable random sources or expanding the randomness of random sources, the SRAM PUF based on the minimum data retention voltage (Data Retention Voltage, DRV) detection lacks an effective DRV detection mechanism and is not very practical; and accelerated aging increases delay loss. The matching method requires a long test time and reduces the service life of the chip; sampling the external high voltage to artificially increase the Vth mismatch of the SRAM input tube will attenuate the statistics of the PUF and increase the test time
In addition, although the digital PUF based on the random on-off of metal wires and contact holes has perfect voltage and temperature stability, it loses the ability of PUF itself to prevent intrusive direct reverse engineering attacks.
[0005] In terms of proposing PUF structures based on new devices, PUFs based on resistive random access memory (Resitive Random Access Memory, RRAM) and magnetic random access memory (Resitive Random Access Memory, RRAM) have high voltage and temperature stability, but they It is not fully compatible with the standard CMOS process, and still needs to use a special process to post-process the chip, and the cost is high
For the method of post-production adjustment and stability improvement, the method of combining voting by show of hands and Error Correction Code (ECC) can obtain a bit error rate close to 0, but it requires a relatively complex digital processing circuit and a large hardware overhead. , and the PUF takes a long time to calibrate every time it starts
In addition, ECC requires on-chip non-volatile memory to store helper data, further increasing hardware overhead
For the method of dynamically adjusting the screening threshold according to the actual test value, a more complex digital processing circuit is required, and the test time is longer. At the same time, many units that do not meet the threshold requirements are discarded, which increases the cost.
[0006] In summary, if the specific circuit of PUF is purely compensated or optimized, its stability cannot be substantially improved.
The reason is that the noise, oscillation frequency, delay, and mismatch between threshold voltages sampled by it are easily affected by voltage and temperature, so that the lowest bit error rate is only 2.5%, and it will also bring practical Low reliability, increased test cost, susceptible to process influence, poor statistical characteristics, etc.
However, digital PUF, RRAM, and MRAM PUF have the problems of being unable to prevent direct reverse engineering attacks and process compatibility.
In addition, the digital circuit overhead of the existing post-trimming method is large, and the test time or PUF preparation time is too long, the cost is high, and the application is limited
[0007] In summary, it can be seen that the existing PUF key stability correction enhancement circuit has the problems of low stability improvement and high hardware cost. It is urgent to find a simple and efficient stability enhancement circuit that can completely eliminate unstable units.