In modern deep-scaled CMOS technologies, various silicon-related pitfalls present challenges to the long-term performance of microprocessors. Such challenges include (1) local hot spots, which breach the thermal limitations of a microprocessor, and (2) transistor aging, especially NBTI, which degrades transistor threshold voltage, ultimately threatening the reliability of the entire memory block. In previous systems, the dummy circuit was placed next to the subject, where the dummy was frequently analyzed, and the readout was used to infer the condition of the target. Due to rapidly changing ambient conditions (e.g., temperature and voltage) and the potential scale of the target dimensions, such metrics may not accurately represent the condition of the target. Moreover, such temperature sensors and canary circuits occupy a significant area.
Therefore, it would be highly preferable to monitor the target circuit in-situ, i.e., to sense the precise transistor at operation. It is also important to achieve an accurate sensing metric. When the temperature is analyzed, the readout should account for voltage and process variations. While sensing the aging degradation, the readout should account for voltage and temperature fluctuations. This would allow testing during in-field operation, while the circuits achieve area-efficiency.
This research had two stages. One result of the first stage was a silicon test chip that was a compact temperature sensor. It involved a family of PTAT+CTAT sensor front-ends that unitized only 6 to 8 conventional CMOS logic devices, yielding a smaller sized chip. The sensor demonstrates accuracy within the target and achieves a 14.3x smaller foot print than preceding published designs. The second product of the first stage was a PMOS aging sensor used in 6T SRAM circuits. The test chip has a real SRAM array, integrated with the proposed PMOS NBTI sensor. It can sense real PMOS NBTI effects in any bit cell (in-situ) and provide robust readings of temperature and voltage (in-field). Intensive aging tests validated the proposed sensing technique.
The second stage was focused on implementing the in-situ and in-field sensing techniques in a real processor. The MIPS microprocessor had a modified instruction cache (I$) and instruction set architecture. With the addition of new instruction aging sensing and minor modification of the circuits, the processor can execute aging sensing opportunistically to evaluate the aging level of its instruction cache. A software framework was developed and verified to estimate the retention voltage of the instruction cache over the lifetime of the chip.
An area-efficient SoC was developed that could transform the instruction cache into an ambient temperature sensor. It had a physically unclonable function (PUF), and it was built with an area-saving technique similar to the earlier work.
This thesis has four chapters. They are presented in chronological and they are aligned with the research described above.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/x6tv-yf89 |
| Date | January 2022 |
| Creators | Yang, Teng |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0017 seconds