4 June 2025 in IEEE Sensors Journal
This paper presents an event-driven, high-speed readout integrated circuit (ROIC) with in-sensor trace detection capabilities for cryogenic infrared focal plane arrays (FPAs), aimed at the next-generation smart infrared motion perception applications. To achieve high accuracy and agile motion recognition with low power consumption and bandwidth requirements, we propose an event-driven image and trace detection method. The proposed pixel consists of two key components: 1) event detection and 2) imaging and trace detection. In response to the demands of cryogenic infrared FPAs, we introduce a synchronized linear response event detection circuit that encodes temporal variations in infrared radiation into 1.5-bit events. Additionally, we propose an in-sensor trace generation method that fuses spatial and temporal information streams, enabling high-accuracy motion recognition while reducing computational and power requirements. When coupled with a post-processor, the event-driven mechanism reduces the need for extensive motion localization and imaging efforts in motion analysis. Meanwhile, the trace information generated by the imaging and trace detection module further enhances motion recognition accuracy for identified objects of interest. The ROIC provides a spatiotemporal redundancy compression and an energy-efficient solution for infrared motion perception applications. A prototype based on a 16×16 array with a 20 μm pixel pitch, fabricated using a 0.18-μm 1P5M CMOS process, demonstrates event-driven motion perception, consuming 11.93 mW (40.75 μW for the array, 10.95 mW for the output buffers) at a 12.2 kHz event and image frame frequency. Detailed test and measurement results of the prototype are provided.
3 March 2026 in IEEE Transactions on Circuits and Systems I: Regular Papers
This paper presents a 16-bit, energy-efficient, fully dynamic successive-approximation-register (SAR) capacitance-to-digital converter (CDC) for single-ended capacitive sensors. A system-level correlated-double-sampling (CDS) scheme is proposed to suppress errors arising from comparator offset, low-frequency noise, and external interference. The delay time characteristics of the inverter-based amplifier (FIA) comparator are characterized for the first time, and an optimal delay time is determined through simulation. An LSB-repetition scheme improves low-noise performance. Dual digital-supply technique reduces the power consumption of the digital circuitry and also extend the delay time of digital delay elements with minimal power overhead, thereby improving energy efficiency. Fabricated in a 180-nm CMOS process, the CDC is validated using a capacitive MEMS pressure sensor, achieving a best resolution of 0.27 fF and a signal-to-noise ratio (SNR) of 82.6 dB among SAR CDCs. Consuming $2.85~\mu $ W with a $56~\mu $ s conversion time, it achieves a Walden figure-of-merit ( $FoM_{w}$ ) of 15 fJ/step. The CDC attains a DNL of -0.76/+0.85 LSB and an INL of -9.16/+9.15 LSB without calibration. Experimental results demonstrate more than $16\times $ improvement in interference rejection at 100 Hz and more than $7\times $ improvement under 1000 Hz, compared to a non-CDS implementation.
13 May 2026 in IEEE Transactions on Circuits and Systems I: Regular Papers
This paper presents an event-driven, dual-mode cryogenic infrared vision sensor that features a two-step time-encoding (TE) analog-to-digital converter (ADC) and a high-speed frame difference (FD) for motion detection. To enable always-on scene-adaptive motion perception with power optimization, the vision sensor is reconfigurable, supporting two operation modes: 1) Precise Imaging (PIM) and 2) Motion Detection (MD). In PIM mode, a two-step TE ADC is proposed for small-pixel-pitch infrared focal plane arrays (IRFPAs). The coarse conversion utilizes pulse frequency modulation (PFM) with an area-efficient true single-phase clock (TSPC) counter. This PFM employs a non-periodic-reset architecture to enable multi-subframe fusion and fine conversion of time-to-first spike (TFS), thereby improving the signal-to-noise ratio (SNR) within a limited pixel area. In MD mode, a high-speed FD-based 1.5-bit event detection is implemented, using subframe coarse conversion without incurring additional memory cost or frame-rate loss. Compared with PIM mode, MD mode reduces power consumption by approximately 40% by deactivating fine conversion and partial output channels. A $1280\times 1024$ prototype with a 15- $\mu $ m pixel pitch was fabricated using a stacking process, in which the digital readout integrated circuit (DROIC) was implemented in a 0.18- $\mu $ m CMOS process and interconnected with the HgCdTe IRFPA via indium bumps. In PIM mode, the prototype achieves 83.2dB SNR with a power consumption of 115.4 mW at 30.5 frames per second (fps) (1.49 pJ $\cdot \mu $ m2/step). In MD mode, with an 8-bit coarse conversion for event detection, the sensor consumes 58.83 mW at 976.6 fps (3.5 fJ/pixel-event), achieving the best event FoM among reported event-driven infrared vision sensors.
1 January 2026 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
1 January 2026 in IEEE Journal of Solid-State Circuits