1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2944236, IEEE Sensors Journal IEEE SENSORS JOURNAL, VOL. XX, NO. X, MAY 2019 1 A Digital Readout IC for Microbolometer Imagers Offering Low Power and Improved Self-Heating Compensation Shahbaz Abbasi, Atia Shafique, Omer Ceylan, Yasar Gurbuz, Member, IEEE Abstract—This paper presents a novel and power efficient readout architecture for uncooled microbolometers that offers adequate noise performance along with a circuit-based method for self-heating compensation. The proposed method employs an event-generation scheme and a time-mode readout chain to achieve dynamic mode of operation resulting in low power. The readout chain consists of a capacitive transimpedance amplifier (CTIA) based current-to-time converter followed by a two stage time-to-digital converter (TDC). The CTIA employs a novel integration scheme for time amplification along with a modified reset mechanism to achieve self-heating compensation. We also present a model to analyze the implications of time-mode readout on imager operation. An experimental readout chip, based on this approach, has been designed using a 130 nm bulk CMOS technology. The proposed architecture offers robust frontend processing and achieves a per channel power consumption of 66 μW , which is considerably lower than the most recently reported designs, while maintaining better than 10-mK readout noise equivalent temperature difference (NETD). Index Terms—Digital readout integrated circuit, uncooled microbolometer, noise equivalent temperature difference (NETD), time-to-digital converter (TDC), self-heating. I. I NTRODUCTION U NCOOLED thermal infrared (IR) imaging systems offer a low-cost alternative to their cooled counterparts. Such imaging systems have been under focus for medical, military and industrial applications [1], [2]. These systems have small pixel pitches and require power efficiency, low noise equivalent temperature difference (NETD) and adequate dynamic range. To achieve low NETD, excellent readout noise performance is needed which requires sensitive analog conditioning cir- cuits in conventional readout architectures. This inevitably necessitates the use of power consuming buffers. Moreover, the commonly found self-heating problem in microbolome- ter imagers has severe artifacts on readout dynamic range. The excess temperature rise caused by self-heating, imposes unnecesarily high precision requirements, thus complicating the readout design. The conventional methods employed to counter this problem, usually require the fabrication of a set of carefully designed reference microbolometers along with complex calibration schemes. This results in added architec- tural complexity. Thus re-examination of these fundamental performance-limiting issues is highly essential to meet the demands of current thermal imaging systems. The authors are with the Faculty of Engineering and Natural Sciences, Sabanci University, Istanbul, Turkey e-mail: yasar@sabanciuniv.edu Manuscript received May, 2019. To reduce readout noise in a power efficient manner, various methods have been proposed. The ROIC in [3] uses a chain of programmable gain/chopper analog amplifiers to cancel offset and noise and, as a result, consumes large power (420 mW for a 320×240 array in a 500-nm CMOS process). A readout channel proposed in [4] and using the same technology node demonstrates 81 mK NETD but consumes 175 mW. The design in [5] employs multiple reference microbolometers to reduce substrate temperature variation and self heating but the readout channel remains analog. Finally, the most recent work in [6] uses digital correlated double sampling (CDS) along with custom designed reference cells to counter offset and gain errors. It reports a power consumption of 45 mW in 350-nm for an 80×60 array with an NETD of 100 mK. Thus with a conventional analog readout architecture, large power consumption seems inevitable if noise performance is to be improved. On the self-heating front, various approaches have been reported for the compensation of its effects. One method uses an external circuitry which emulates the change in the detector voltage for compensation purposes [7]. This method is simple but causes excessive noise coupling. A more commonly used method is based on the use of an identical reference mi- crobolometer that undergoes a similar rise in resistance caused by self-heating and can therefore be used to cancel it out [8]. Perfect cancellation of self-heating induced temperature change is not possible using this method owing to imperfect detector matching. A further cause of the ineffectiveness of this method is the fact that the reference detectors have higher thermal conductance compared to the active detectors. Reference detectors are designed this way so that they can have faster cooling cycle and can, therefore, be addressed more frequently serving a whole column turn-by-turn within a frame. This additional mismatch makes it difficult to achieve complete cancellation. Another method corrects self-heating induced artifacts by applying an equivalent correction signal to the detector output before the signal is amplified or integrated [9]. To address the power consumption and self-heating con- cerns discussed above, a dynamic and highly digital readout chain based on time-mode processing is proposed in this work. Introducing the concept of dynamic event generation into frame synchronous readouts provides an opportunity to obtain adequate noise performance in a power efficient manner. By converting the microbolometer temperature to a time-mode signal, using digital events, it is possible to