On-die parameter extraction from path-delay measurements

Abstract

Device-parameter estimation through path-delay measurement, which facilitates fast on-die performance prediction and diagnosis, is proposed. With the proposed technique, delays of a set of paths consisting of different logic cells are monitored. Based on the pre-characterized parameter to delay sensitivity, the process variation of a chip is estimated as an inverse problem. Discussion of desirable logic cell combination to form paths that maximize estimation accuracy is presented. Measurement of ring oscillator arrays composed of standard and customized logic cells resulted in consistent estimation of threshold voltages. Measurement accuracy is greatly enhanced by the proposed good logic cell combinations.

Path clustering for adaptive test

“…The clusters are approximately symmetric between FS (fastslow) and SF (slow-fast) conditions because the effect of increasing/decreasing threshold voltages of nMOS and pMOS transistors result in similar speed-down/speed-up of the path delay. However, due to the different sensitivity of the logic cells to nMOS and pMOS threshold voltages break its symmetry, which has been experimentally observed in [9]. In this case, the numbers of test paths in FF and SS conditions are 33 and 6027, respectively.…”

“…Here, the effectiveness of the adaptive test is affected by the accuracy of the sensor circuits and path-clustering algorithm. In [9], the parameter extraction method from ring oscillators' frequencies, which is suitable for the adaptive test, is proposed. It correctly estimates process condition of each chip.…”

“…In the adaptive test, the test path sets that are the most suitable for respective process conditions are prepared beforehand, and selected on-the-fly according to the chip performance observation. For that purpose, the sensor circuit [9] is embedded on a chip or on a wafer, with which the chip's process condition is identified.…”

Low-cost sensing with ring oscillator arrays for healthier reconfigurable systems

“…The coefficient of variation across different oscillator instances due to this variation is related to the number of stages by [Takahashi et al 2009]:…”