Predicting circuit aging using ring oscillators

Abstract

Abstract-This paper presents a method for inferring circuit delay shifts due to bias temperature instability using ring oscillator (ROSC) sensors. This procedure is based on presilicon analysis, postsilicon ROSC measurements, a new aging analysis model called the Upperbound on f M ax (UofM), and a look-up table that stores a precomputed degradation ratio that translates delay shifts in the ROSC to those in the circuits. This method not only yields delay estimates within 0.2% of the true values with very low runtime, but is also independent of temperature and supply voltage variations.

ReSCALE: Recalibrating sensor circuits for aging and lifetime estimation under BTI

“…Between measurement instants, we estimate delay degradation in the CUT by measuring the delay degradation in the ROSC and multiplying it by D. At time zero, we use the worst-case aging curve to obtain D as in [15]. This value has been shown to be constant under process, voltage, and temperature variations.…”

“…Further, D is recalibrated and updated, if necessary, in the LUT. Overhead: The overheads of this scheme are due to (a) Area: this overhead is low, particularly for large circuits, since a single ROSC can be used for aging prediction in multiple nearby circuits, as quantified in [15]. (b) CUT delay: the additional load due to the true delay measurement circuitry is negligible and does not significantly affect path delays as demonstrated in [13].…”

“…4. A simpler and accurate representation for the CUT delay, which we use in this work, was introduced in [15] using a tight upperbound on DCUT (t), denoted by D U of M (t) and shown in Fig. 4.…”

“…By placing the ROSC close enough to the CUT, it may experience similar process variations, and undergo similar temperature and voltage stress, as the CUT. This is shown in [15] to keep the value of D constant over such variations.…”

“…At each measurement instant, the CUT delay is measured, providing an accurate view of the actual stressing conditions that it experiences, and D is recalibrated. The LUT is then updated with the measured delay and the new D. We now explain the theory behind the recalibration procedure, developing a new result in Theorem 1 that generalizes the presilicon upper bound of [15] to our postsilicon approach based on measurement and recalibration.…”