15 July 2019 in Nature Electronics
4 May 2020 in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
Analog compute-in-memory with resistive random access memory (RRAM) devices promises to overcome the data movement bottleneck in data-intensive AI and machine learning. RRAM crossbar arrays improve the efficiency of vector-matrix multiplications (VMM), which is a vital operation in these applications. The prototype IC is the first complete, fully-integrated analog-RRAM CMOS coprocessor. This article focuses on the digital and analog circuitry that supports efficient and flexible RRAM-based computation. A passive 54x108 RRAM crossbar array performs VMM in the analog domain. Specialized mixed-signal circuits stimulate and read the outputs of the RRAM crossbar. The single-chip CMOS prototype includes a RISC processor interfaced to a memory-mapped mixed-signal core. In the mixed-signal core, ADCs and DACs interface to the passive RRAM crossbar. The RISC processor controls the mixed-signal circuits and the algorithm data path. The system is fully programmable and supports forward and backward propagation. As proof of concept, a fully-integrated 0.18μm CMOS prototype with a post-processed RRAM array demonstrates several key functions of machine learning including online learning. The mixed-signal core consumes 64mW at an operating frequency of 148MHz. The total system power consumption, considering the mixed-signal circuitry, the digital processor, and the passive RRAM array is 307mW. The maximum theoretical throughput is 2.6 GOPS at an efficiency of 8.5 GOPS/W.
9 October 2024 in IEEE Journal of Solid-State Circuits
In this article, we introduce a new dynamically biased ring amplifier that is tolerant to mismatch and PVT variation without requiring bias calibration, and we verify it in a 12-bit 400-MS/s pipelined-SAR analog-to-digital converter (ADC), fabricated in an 8-nm FinFET process. Our novel ring amplifier solves the biasing issues inherent in conventional ring amplifiers while maintaining the benefits of high gain, slew-based charging, and nearly rail-to-rail output swing. We also propose a technique to enhance the DC accuracy of a switched-capacitor common-mode feedback (CMFB) without consuming additional power, which we named feedback voltage sampling CMFB. Furthermore, we introduce a full-scale matching residue amplification technique for the prototype pipelined-SAR ADC to utilize the top-plate input sampling for the first-stage SAR ADC, resulting in faster and lower power conversion. The prototype ADC demonstrates the robustness of our dynamically biased ring amplifier to mismatch and PVT variation without any interstage gain, bias, or reference calibration, and achieves 64.4-dB SNDR and 77.6-dB SFDR for a low-frequency input while consuming 2.08 mW. This measured performance is equivalent to Walden and Schreier FoMs of 3.8 fJ/conversion $\cdot$ step and 174.2 dB, respectively.
6 June 2024 in IEEE Journal of Solid-State Circuits
This article presents an intrinsically linear continuous-time (CT) zoom analog-to-digital converter (ADC) architecture that uses a coarse first-order CT delta–sigma modulator ( $\Delta\Sigma$ M) followed by a fine third-order CT- $\Delta\Sigma$ M. The 1.5-bit output of the coarse CT- $\Delta\Sigma$ M is filtered by an 8-tap finite impulse response (FIR) filter, and then, a current-steering DAC provides intrinsically linear outputs to the fine $\Delta\Sigma$ M. The associated delay is mitigated by increasing the coarse sampling rate by a factor of 4. The coarse $\Delta\Sigma$ M significantly relaxes the requirements of the fine $\Delta\Sigma$ M, but despite the faster sampling rate, consumes negligible power, because its noise is canceled for the final output, allowing the use of a relatively large input resistor. The ADC prototype fabricated in a 65-nm CMOS process occupies an area of 0.124 $\text{mm}^\text{2}$ and consumes only 243 $\mu$ W from a 1.2-V supply. It achieves 92.0-dB signal-to-noise and distortion ratio (SNDR), 92.3-dB signal-to-noise ratio (SNR), 97.4-dB dynamic range (DR), and 105.7-dB spurious-free DR (SFDR) over a 50-kHz bandwidth without using DAC linearization or calibration. This corresponds to Schreier FoMs of 175.1 dB (SNDR) and 180.5 dB (DR).