Multi-Valued Physical Unclonable Functions

点击翻译按钮获取中文摘要

Since two decades, the level of semiconductor manufacturing went below the mark of 100 nm CMOS nodes. Thereby, a drastically increasing effect is the influence of process variation manifesting as device mismatch. However, it was also discovered that non-predictable and irreproducible device mismatch can be utilized to develop specific types of circuits for hardware-based chip identification, called Physical Unclonable Functions (PUFs). The main purpose of PUFs is to generate unique keys or fingerprints from these physical properties with non-predictable variations. In order for a specific instance to have a response stable over a wide range of environmental conditions, error correction (channel coding) methods have to be applied.Even though this fingerprint is based on the non-ideal manufacturing-induced mismatch of identically designed components, and this mismatch follows statistical distributions, still, in the vast majority of today's PUFs, the readout is simply binary. However, for the purpose of PUFs and in view of finite block length, this is not prescribed or unavoidable and may even be not the optimum approach w.r.t. the entire system. Thus, multi-valued PUFs (MPUFs) should be considered, where the fingerprint is a word with elements drawn from an alphabet with (finite) cardinality greater than two. Utilizing the true analog distributions of the hardware and mapping it into multi-valued, non-binary responses could provide a more diverse and distinguishable behavior of the PUF circuit yielding a much greater entropy of the response.The proposed project combines the expertise of integrated circuit design and communication theory with the goal to exploit the multi-valued nature of the PUF source of randomness, to increase the entropy of its readout and to generate robust readouts. The complexity tradeoffs between implementation effort of enhanced channel and source coding versus enhanced circuit design measure will be analyzed.To this end, on the one hand, by using integrated circuit design and simulation, realistic model building and model verification is performed. Suited multi-valued PUF sources will be selected and optimized. On the other hand, source coding and coded modulation schemes adapted to themulti-valued nature will be designed. Both parts strongly interact: the coding part will provide requirements and goal metrics for enhanced MPUFs, which will lead to improved MPUF unit cells; the realistic hardware models will be the basis for adapted, highly performing coding schemes.The outcome of the project is expected to be a new class of physical unclonable functions with unseen statistical properties and reliability for fingerprints, authentication or key generation/key storage for cryptography.