On-Chip Sensors for Process, Aging, and Temperature Variation Kaushik Mazumdar Jared Robertson
Overview Motivation Existing Methods Test Circuit Simulations Summary References
Motivation Local Process Variation Global Process Variation Temperature Variation NBTI/PBTI Degradation [1]
Existing Methods Device Array Ring Oscillator Large Area Long Test Time Many I/O Pins Ring Oscillator Doesn’t Measure Mismatch or Variation Gradients Can’t Measure PMOS/NMOS Variation Separately [2]
Weak Pull-down / Strong Pull-up Strong Pull-down / Weak Pull-up Sensor Circuit Virtual Gnd Virtual Vdd Weak Pull-down / Strong Pull-up 203.5 mV 1.091 V Strong Pull-down / Weak Pull-up 25 mV 965.4 mV [3]
Simulation Used Monte Carlo to randomly vary Vt in transistors PMOS 1 PMOS 2 PMOS 3 PMOS 4 PMOS 5 Process 1 3.685 4.0063 4.2199 4.286 4.216 Process 2 3.5464 3.807 4.0457 4.0708 4.0807 Process 3 3.5092 3.7241 4.0478 3.9348 4.0488 Process 4 3.4925 3.7479 4.0309 3.9846 4.0536 Process 5 3.4494 3.7377 3.9398 3.9752 4.027 Used Monte Carlo to randomly vary Vt in transistors Extracted frequency of ring oscillator for each active header/footer over various process runs Extracted statistical data from frequency values
Threshold Voltage Extraction Random variability and systematic shift can be measured from data Use equation derived from simulation to calculate device-level variations with oscillator frequency
Sensitivity Knobs Circuit is more sensitive to process variation when header/footer sizing is minimized Circuit becomes more sensitive to process variation as number of oscillator stages increases
To Maximize Process Variation Sensitivity Sensitivity Knobs Circuit is more sensitive to process variation at higher Vdd Set knobs to maximize sensitivity for process variation Placing many sensors around the chip exploits the increased sensitivity for fewer ring oscillator stages To Maximize Process Variation Sensitivity Vdd # Stages Header/Footer Sizing High Low Minimum
Process Variation Sensitivity Thermal Sensor The oscillator can be stretched across a chip to sense temperature variation Process variation sensitivity should be minimized Our simulations achieved a sensitivity reduction of 10x Process Variation Sensitivity Vdd # Stages Header Sizing Standard Deviation Maximum 1.1 11 90nm 0.1281 Minimum 0.7 23 300nm 0.0126
NBTI/PBTI Sensor Frequency Sensor Measures Beat Frequency Differential Measurement Method Isolates NBTI/PBTI Degradation from other variation sources [4]
Odometer Output
Summary Analyzed variation issues and existing solutions Showed simulation methodology and results Tuned oscillator circuit to isolate process variation from temperature Displayed how circuit can be used as a temperature sensor Developed NBTI/PBTI degradation detection method Discussion of possible future work
References M. A. Alam, S. Mahapatra, "A comprehensive model of PMOS NBTI degradation," Microelectronics Reliability, vol. 45, pp. 71--81, 2005. Kanak Agarwal; Kanak Sani Nassif; “Characterizing process variation in nanometer CMOS”, Proceedings of the 44th annual conference on Design automation, June 04-08, 2007, San Diego, California. Agarwal, Kanak, “On-die sensors for measuring process and environmental variation in integrated circuits.” IBM Corps. Austin, TX. 2010. Tae-Hyoung Kim; Persaud, R.; Kim, C.H.; "Silicon Odometer: An On-Chip Reliability Monitor for Measuring Frequency Degradation of Digital Circuits," Solid-State Circuits, IEEE Journal of , vol. 43, no. 4, pp. 874-880, April 2008. Wei Zhao; Liu, F.; Agarwal, K.; Acharyya, D.; Nassif, S.R.; Nowka, K.J.; Yu Cao; , "Rigorous Extraction of Process Variations for 65-nm CMOS Design," Semiconductor Manufacturing, IEEE Transactions on , vol.22, no.1, pp.196-203, Feb. 2009. Kumar, R.; Kursun, V.; , "Temperature Variation Insensitive Energy Efficient CMOS Circuits in a 65nm CMOS Technology," Circuits and Systems, 2006. MWSCAS '06. 49th IEEE International Midwest Symposium on , vol.2, no., pp.226-230, 6-9 Aug. 2006.