2018 Theses Doctoral
Very-Large-Scale-Integration Circuit Techniques in Internet-of-Things Applications
Heading towards the era of Internet-of-things (IoT) means both opportunity and challenge for the circuit-design community. In a system where billions of devices are equipped with the ability to sense, compute, communicate with each other and perform tasks in a coordinated manner, security and power management are among the most critical challenges.
Physically unclonable function (PUF) emerges as an important security primitive in hardware-security applications; it provides an object-specific physical identifier hidden within the intrinsic device variations, which is hard to expose and reproduce by adversaries. Yet, designing a compact PUF robust to noise, temperature and voltage remains a challenge.
This thesis presents a novel PUF design approach based on a pair of ultra-compact analog circuits whose output is proportional to absolute temperature. The proposed approach is demonstrated through two works: (1) an ultra-compact and robust PUF based on voltage-compensated proportional-to-absolute-temperature voltage generators that occupies 8.3× less area than the previous work with the similar robustness and twice the robustness of the previously most compact PUF design and (2) a technique to transform a 6T-SRAM array into a robust analog PUF with minimal overhead. In this work, similar circuit topology is used to transform a preexisting on-chip SRAM into a PUF, which further reduces the area in (1) with no robustness penalty.
In this thesis, we also explore techniques for power management circuit design.
Energy harvesting is an essential functionality in an IoT sensor node, where battery replacement is cost-prohibitive or impractical. Yet, existing energy-harvesting power management units (EH PMU) suffer from efficiency loss in the two-step voltage conversion: harvester-to-battery and battery-to-load. We propose an EH PMU architecture with hybrid energy storage, where a capacitor is introduced in addition to the battery to serve as an intermediate energy buffer to minimize the battery involvement in the system energy flow. Test-case measurements show as much as a 2.2× improvement in the end-to-end energy efficiency.
In contrast, with the drastically reduced power consumption of IoT nodes that operates in the sub-threshold regime, adaptive dynamic voltage scaling (DVS) for supply-voltage margin removal, fully on-chip integration and high power conversion efficiency (PCE) are required in PMU designs. We present a PMU–load co-design based on a fully integrated switched-capacitor DC-DC converter (SC-DC) and hybrid error/replica-based regulation for a fully digital PMU control. The PMU is integrated with a neural spike processor (NSP) that achieves a record-low power consumption of 0.61 µW for 96 channels. A tunable replica circuit is added to assist the error regulation and prevent loss of regulation. With automatic energy-robustness co-optimization, the PMU can set the SC-DC’s optimal conversion ratio and switching frequency. The PMU achieves a PCE of 77.7% (72.2%) at VIN = 0.6 V (1 V) and at the NSP’s margin-free operating point.
- Li_columbia_0054D_14964.pdf application/pdf 2.73 MB Download File
More About This Work
- Academic Units
- Electrical Engineering
- Thesis Advisors
- Seok, Mingoo
- Ph.D., Columbia University
- Published Here
- October 18, 2018