The unclonability of PUFs makes PUFs suitable to be the root of trust of IoT devices. However, if some attackers try to rebuild the PUF model by deep learning, they can get most of the CRPs.Therefore, we should design the protection mechanisms for the PUF from the attackers.On the other hand, when executing intelligent calculations on IoT devices, by using compute-in-memory (CIM) architecture, we avoid the latency and power consumption of memory access and increase the accelerator＇s overall throughput.
Our team designs multi-layer protection mechanism to make sure the system resists the attacks. First, we imbedded the PUF into the system so that it can’t be accessed directly from the outside. Then we designed the protection mechanism against side channel attacks. Finally, we used authentication and memory protection mechanism to check if the user’s programs are verified and invalidate illegal access to the confidential information, so that we can guarantee the security of this device.

1.Our team makes sure the security of the IoT device’s root of trust and CIM’s data through multi-layer hardware security protection mechanisms.
2.The CIM circuit our team developed for multi-bit weight and multi-bit input data uses the methodologies of time-domain quantization and weight bit-partitioned accumulation to increase the energy efficiency and accuracy.

Our team takes the PUF as the new proposal of root of trust, developing PUF protection mechanism, and based on the root of trust, designing a device which has multi-layer protection mechanisms and advanced accelerator containing CIM. The technology of our research results can be transferred to interested companies, providing domestic industry proper solutions.