A 4.82-µW 183.4dB-FoMSNDR CT Incremental Tracking-Zoom Sensor Readout Frontend with Floating-Gm-CCO Integrator

Emerging edge applications in IoT, medical, and wearables often demand continuous sensor monitoring with limited battery or energy harvester power. Direct sensor readout frontends are highly preferred, as they significantly simplify signal conditioning circuits and reduce system costs [1]–[7]. Edge applications require sensor readout circuits to offer a low input-referred noise $(\leq 10\mu \mathrm{V}_{\text{rms}})$ and a medium conversion speed (≥10kHz). To process small varying signals with the presence of large potential interference and artifacts around 100mV range, the readout frontend should offer a large dynamic range (DR)≥80dB. Conventional $\Delta\Sigma Ms$ can realize such performance while requiring either high-order $\Delta\Sigma$ loops or large OSR [1]–[6], limiting the energy efficiency. The zoom architecture, combining energy-efficient SAR with high-resolution $\Delta\Sigma M$, shows great potential for edge signal readout. With most of the input quantized by the coarse SAR, the $\Delta\Sigma$ loop only needs to handle a small residue, and thus, the zoom architecture offers a large input range, high linearity, and low noise simultaneously [8]–[10]. Nevertheless, during conversion, signals with a large variation can move beyond the fine quantization range. Dynamic zoom performs coarse SAR quantization at each cycle to ensure the input stays within the fine quantization range, but the frequent DAC switching limits its energy efficiency [11]. To alleviate this issue, recent developments in sensor recordings have adopted range tracking techniques. By estimating input value with a dedicated loop-filter monitor [7] or a post-decimation estimator [12] and updating the fine quantization range accordingly, signals can be kept within the quantization range without running coarse quantizing in each fine cycle. Yet, the former adds extra always-on analog comparators and also requires one SAR conversion per zoom cycle to accommodate system-level chopping. The latter relies on complicated logics and large inter-stage redundancy to extend the acceptable slew rate. As a result, both approaches incur significant extra energy and area cost.