Shuo Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Shuo Li has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
- Benton H. Calhoun
Department of Electrical and Computer Engineering, University of Virginia, Charlottesville, VA, USA
5 shared publications - Jacob Breiholz
University of Virginia
2 shared publications - Xinjian Liu
Department of Electrical and Computer Engineering, University of Virginia, Charlottesville, VA, USA
2 shared publications Abhishek Roy
Infrastructure Processor Business Unit, Marvell Semiconductor, Inc., Marlborough, USA
1 shared publication- Abhishek Roy
University of Virginia, Charlottesville, VA 22903 USA
1 shared publication - Avish Kosari
University of Michigan, Ann Arbor, MI, USA
1 shared publication
Latest Publications
- 28 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 73 (2) , 1010-1023
Characterizing the Intrinsic Bank-Level Accuracy Versus Energy Trade-Off of SRAM-Based Analog In-Memory Computing Architectures in 28 nm CMOS
In-memory computing (IMC) architectures are emerging as the next-generation computing platforms for deep learning applications, surpassing traditional von Neumann architectures due to their superior energy efficiency and computational density. However, there is a lack of comprehensive understanding of the trade-off between accuracy at the bank level and energy consumption of SRAM-based IMCs, which execute computations in the analog domain. To gain a complete understanding of IMC’s fundamental limits and design space, we first adopt the signal-to-noise-and-distortion ratio (SNDR) as a metric, inspired by ADC design, to quantify the accuracy at the bank level. Second, we propose a methodology for accurately measuring SNDR. Finally, we propose a chip architecture for to directly measure pre-ADC SNDR. This characterization chip, incorporating mainstream IMC architectures, is fabricated in 28 nm CMOS technology. Measurement results indicate that SNDR and energy consumption are trade-offs that can be adjusted through the dot-product dimension and wordline voltage for charge/current summing models, and through input-driver voltage for a charge-redistribution model. To the best of our knowledge, this is the first experimental quantification of the trade-off between bank-level accuracy and energy consumption for SRAM-based IMC architectures through silicon measurements. These results underscore the importance of characterizing IMC SNDR to properly benchmark and compare future IMC designs.
- 28 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 73 (2) , 1010-1023
Characterizing the Intrinsic Bank-Level Accuracy Versus Energy Trade-Off of SRAM-Based Analog In-Memory Computing Architectures in 28 nm CMOS
In-memory computing (IMC) architectures are emerging as the next-generation computing platforms for deep learning applications, surpassing traditional von Neumann architectures due to their superior energy efficiency and computational density. However, there is a lack of comprehensive understanding of the trade-off between accuracy at the bank level and energy consumption of SRAM-based IMCs, which execute computations in the analog domain. To gain a complete understanding of IMC’s fundamental limits and design space, we first adopt the signal-to-noise-and-distortion ratio (SNDR) as a metric, inspired by ADC design, to quantify the accuracy at the bank level. Second, we propose a methodology for accurately measuring SNDR. Finally, we propose a chip architecture for to directly measure pre-ADC SNDR. This characterization chip, incorporating mainstream IMC architectures, is fabricated in 28 nm CMOS technology. Measurement results indicate that SNDR and energy consumption are trade-offs that can be adjusted through the dot-product dimension and wordline voltage for charge/current summing models, and through input-driver voltage for a charge-redistribution model. To the best of our knowledge, this is the first experimental quantification of the trade-off between bank-level accuracy and energy consumption for SRAM-based IMC architectures through silicon measurements. These results underscore the importance of characterizing IMC SNDR to properly benchmark and compare future IMC designs.
- 9 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 45 (4) , 1691-1703
DAMIL-DCIM+: Automated Dataflow-Aware Layout Synthesis for Digital CIM With Self-Assembled Bitcell Units and MILP-Based Optimization
Digital computing-in-memory (DCIM) systems integrate complex digital logic with parasitic-sensitive bitcell arrays, presenting unique physical design challenges. Conventional design strategies often fall short in these systems due to irregular dataflow patterns and excessive interconnect lengths, which degrade performance and increase parasitic effects. As a result, current DCIM implementations frequently rely on manual layout, which is both time-consuming and a major bottleneck in the design cycle. While existing DCIM layout synthesis frameworks attempt to automate this process using template-based placement methods inspired by manual design, their rigid constraints can lead to inefficient area utilization and increased core sizes. To address these limitations, we propose DAMIL-DCIM+, a novel placement framework that combines the structural clarity of template-based methods with the flexibility of optimization-based techniques. Specifically, DAMIL-DCIM+ employs a global dataflow-aware floorplan to guide placement and leverages MILP-based detailed placement to optimize wirelength and preserve dataflow regularity. Inspired by self-assembling design principles, this approach enables scalable and structured integration of parasitic-sensitive components. The hybrid methodology of DAMIL-DCIM+ reduces total wirelength, lowers parasitic effects, and enhances performance while maintaining design regularity. Experimental results on a 28-nm DCIM circuit demonstrate that DAMIL-DCIM+ improves operating frequency by 25.2% and reduces power consumption by 19.6% compared to Cadence Innovus, without increasing core area.
- 9 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 45 (4) , 1691-1703
DAMIL-DCIM+: Automated Dataflow-Aware Layout Synthesis for Digital CIM With Self-Assembled Bitcell Units and MILP-Based Optimization
Digital computing-in-memory (DCIM) systems integrate complex digital logic with parasitic-sensitive bitcell arrays, presenting unique physical design challenges. Conventional design strategies often fall short in these systems due to irregular dataflow patterns and excessive interconnect lengths, which degrade performance and increase parasitic effects. As a result, current DCIM implementations frequently rely on manual layout, which is both time-consuming and a major bottleneck in the design cycle. While existing DCIM layout synthesis frameworks attempt to automate this process using template-based placement methods inspired by manual design, their rigid constraints can lead to inefficient area utilization and increased core sizes. To address these limitations, we propose DAMIL-DCIM+, a novel placement framework that combines the structural clarity of template-based methods with the flexibility of optimization-based techniques. Specifically, DAMIL-DCIM+ employs a global dataflow-aware floorplan to guide placement and leverages MILP-based detailed placement to optimize wirelength and preserve dataflow regularity. Inspired by self-assembling design principles, this approach enables scalable and structured integration of parasitic-sensitive components. The hybrid methodology of DAMIL-DCIM+ reduces total wirelength, lowers parasitic effects, and enhances performance while maintaining design regularity. Experimental results on a 28-nm DCIM circuit demonstrate that DAMIL-DCIM+ improves operating frequency by 25.2% and reduces power consumption by 19.6% compared to Cadence Innovus, without increasing core area.
- 19 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (7) , 2272-2285
A Sub-μW Energy-Performance-Aware IoT SoC With a Triple-Mode Power Management Unit for System Performance Scaling, Fast DVFS, and Energy Minimization
This article presents an ultra-low-power (ULP) Internet-of-Things (IoT) system-on-chip (SoC) using a triple-mode power management unit (PMU) to achieve self-adaptive power–performance scaling and energy-minimized operation. The proposed PMU comprises three modes: energy-aware (EA) mode, performance-aware (PA) mode, and minimum energy point (MEP) tracking mode. By controlling a microprocessor with the three modes, the SoC can adaptively scale its frequency and supply voltage based on either the input energy availability or the task priority. To achieve robust and rapid mode transitions, the SoC adopts fast dynamic voltage and frequency scaling (DVFS) and fast load transient response (FLTR) through asynchronous control. For energy-minimized operation, a sub-nW constant-energy-cycle (CEC) algorithm keeps the microprocessor operating at the MEP with a 0.026-mm $^{2}$ area overhead. In addition, the on-chip integration of a bias generator (BG), clock (CLK), and power-on-reset block empowers the SoC to be a fully self-contained system. Fabricated in 65-nm CMOS, measurement results show that the SoC has a minimum power consumption of 194.3 nW at 180 Hz. The proposed PMU achieves 5.2-nW quiescent power and 92.6% peak efficiency while maintaining $>$ 80% efficiency from 190 nW to 3 mW. The MEP tracking (MEPT) circuits achieve $<$ 2.3% energy per cycle error and $<$ 18 mV voltage tracking error. The measured quiescent power of the MEPT circuits in the idle mode is 379 pW, which only accounts for 0.19% of the total system power. Measurements of the triple-mode transitions show that this SoC is well suited for resource-constrained IoT applications.
- 19 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (7) , 2272-2285
A Sub-μW Energy-Performance-Aware IoT SoC With a Triple-Mode Power Management Unit for System Performance Scaling, Fast DVFS, and Energy Minimization
This article presents an ultra-low-power (ULP) Internet-of-Things (IoT) system-on-chip (SoC) using a triple-mode power management unit (PMU) to achieve self-adaptive power–performance scaling and energy-minimized operation. The proposed PMU comprises three modes: energy-aware (EA) mode, performance-aware (PA) mode, and minimum energy point (MEP) tracking mode. By controlling a microprocessor with the three modes, the SoC can adaptively scale its frequency and supply voltage based on either the input energy availability or the task priority. To achieve robust and rapid mode transitions, the SoC adopts fast dynamic voltage and frequency scaling (DVFS) and fast load transient response (FLTR) through asynchronous control. For energy-minimized operation, a sub-nW constant-energy-cycle (CEC) algorithm keeps the microprocessor operating at the MEP with a 0.026-mm $^{2}$ area overhead. In addition, the on-chip integration of a bias generator (BG), clock (CLK), and power-on-reset block empowers the SoC to be a fully self-contained system. Fabricated in 65-nm CMOS, measurement results show that the SoC has a minimum power consumption of 194.3 nW at 180 Hz. The proposed PMU achieves 5.2-nW quiescent power and 92.6% peak efficiency while maintaining $>$ 80% efficiency from 190 nW to 3 mW. The MEP tracking (MEPT) circuits achieve $<$ 2.3% energy per cycle error and $<$ 18 mV voltage tracking error. The measured quiescent power of the MEPT circuits in the idle mode is 379 pW, which only accounts for 0.19% of the total system power. Measurements of the triple-mode transitions show that this SoC is well suited for resource-constrained IoT applications.
- Open Access
- 10 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
A 956pW Switched-Capacitor Sub-Bandgap Reference with 0.44-to-3.3V Supply Range, -67dB PSRR, and 0.2% within-Wafer Inaccuracy for Nanowatt IoT Systems
A document by Shuo Li. Click on the document to view its contents.
- 10 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessA 956pW Switched-Capacitor Sub-Bandgap Reference with 0.44-to-3.3V Supply Range, -67dB PSRR, and 0.2% within-Wafer Inaccuracy for Nanowatt IoT Systems
A document by Shuo Li. Click on the document to view its contents.
- 9 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 32 (1) , 190-194
Scalable All-Analog LDOs With Reduced Input Offset Variability Using Digital Synthesis Flow in 65-nm CMOS
The design and verification process for analog circuits can be long and tedious, wherein designers rely heavily on manual effort to create circuits and draw layouts, thereby limiting turn-around-time and design scale and increasing costs. Various previous works have tried to solve this issue by leveraging digital automated place-and-route (APR) tools, but they involve replacing analog elements with digital counterparts, thereby dampening performance. In this work, we propose a digital flow-based approach to design all-analog circuits that dramatically speeds up the design and layout process while retaining the benefits of true analog topologies and demonstrate the performance for three low-dropout regulators (LDOs). Fabricated in 65-nm CMOS, measurement results show that the generated LDOs achieve up to 99.95% peak current efficiency, a figure-of-merit (FOM) of 4.6 ps, and up to 63.93% reduction in input offset variability with respect to their manually designed counterparts.
- 9 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 32 (1) , 190-194
Scalable All-Analog LDOs With Reduced Input Offset Variability Using Digital Synthesis Flow in 65-nm CMOS
The design and verification process for analog circuits can be long and tedious, wherein designers rely heavily on manual effort to create circuits and draw layouts, thereby limiting turn-around-time and design scale and increasing costs. Various previous works have tried to solve this issue by leveraging digital automated place-and-route (APR) tools, but they involve replacing analog elements with digital counterparts, thereby dampening performance. In this work, we propose a digital flow-based approach to design all-analog circuits that dramatically speeds up the design and layout process while retaining the benefits of true analog topologies and demonstrate the performance for three low-dropout regulators (LDOs). Fabricated in 65-nm CMOS, measurement results show that the generated LDOs achieve up to 99.95% peak current efficiency, a figure-of-merit (FOM) of 4.6 ps, and up to 63.93% reduction in input offset variability with respect to their manually designed counterparts.
- 6 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 57 (7) , 2054-2067
A Sub-nW 93% Peak Efficiency Buck Converter With Wide Dynamic Range, Fast DVFS, and Asynchronous Load-Transient Control
This article presents a highly efficient buck converter with sub-nW quiescent power and wide dynamic range for ultra-low-power (ULP) Internet-of-Things (IoT) systems-on-chip (SoCs). To optimize the SoC power consumption and performance, this buck converter supports fast dynamic voltage and frequency scaling (DVFS) and fast load-transient response (FLTR) using an asynchronous control scheme. To achieve robust and high-efficiency power delivery over process, voltage, and temperature variations, an adaptive deadtime controller (ADTC) is proposed with minimized area and power overhead. The power stage and gate drivers are optimized by a length split technique and a strong-up weak-down (SuWd) scheme to achieve low quiescent power. In addition, the buck converter is fully self-contained with a bias generator (BG), clock, and power-on-reset (PoR) integrated on-chip. Fabricated in 65-nm CMOS, measurement results show that the buck converter achieves 802-pW quiescent power and 93% peak efficiency at 1.5-V input voltage. The measured dynamic range is from 0.5 to 2.75 mW, which is over six orders of magnitude. The measured voltage droop is 56 mV for a 45-nA-to-1-mA load current step and DVFS up- and down-tracking takes 10.57 and $19.81~\mu \text{s}$ for an 88- and 92-mV reference step, respectively. This sub-nW buck converter integrates fast DVFS and FLTR features with a wide dynamic range making it suitable for ULP IoT applications.
- 6 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 57 (7) , 2054-2067
A Sub-nW 93% Peak Efficiency Buck Converter With Wide Dynamic Range, Fast DVFS, and Asynchronous Load-Transient Control
This article presents a highly efficient buck converter with sub-nW quiescent power and wide dynamic range for ultra-low-power (ULP) Internet-of-Things (IoT) systems-on-chip (SoCs). To optimize the SoC power consumption and performance, this buck converter supports fast dynamic voltage and frequency scaling (DVFS) and fast load-transient response (FLTR) using an asynchronous control scheme. To achieve robust and high-efficiency power delivery over process, voltage, and temperature variations, an adaptive deadtime controller (ADTC) is proposed with minimized area and power overhead. The power stage and gate drivers are optimized by a length split technique and a strong-up weak-down (SuWd) scheme to achieve low quiescent power. In addition, the buck converter is fully self-contained with a bias generator (BG), clock, and power-on-reset (PoR) integrated on-chip. Fabricated in 65-nm CMOS, measurement results show that the buck converter achieves 802-pW quiescent power and 93% peak efficiency at 1.5-V input voltage. The measured dynamic range is from 0.5 to 2.75 mW, which is over six orders of magnitude. The measured voltage droop is 56 mV for a 45-nA-to-1-mA load current step and DVFS up- and down-tracking takes 10.57 and $19.81~\mu \text{s}$ for an 88- and 92-mV reference step, respectively. This sub-nW buck converter integrates fast DVFS and FLTR features with a wide dynamic range making it suitable for ULP IoT applications.
- 14 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 56 (4) , 1241-1253
A 0.5-V 560-kHz 18.8-fJ/Cycle On-Chip Oscillator With 96.1-ppm/°C Steady-State Stability Using a Duty-Cycled Digital Frequency-Locked Loop
On-chip oscillators are popular clocking solutions for a wide range of circuits and systems due to their ease of integration and low form factor, but their energy efficiency is typically limited to the pJ/cycle range by a number of contributors, such as active biasing currents, frequency dividers, and comparators. This work presents an on-chip oscillator for energy-efficient Internet-of-Things (IoT) applications based on a duty-cycled digital frequency-locked loop (DFLL) that reduces energy by disabling energy-hungry components and only periodically reactivating them to keep the output frequency stabilized during temperature drifts. A test chip is implemented in 65-nm CMOS and achieves 18.8 fJ/cycle (10.5 nW at 560 kHz) while maintaining an average steady-state temperature stability of 96.1 ppm/°C from 0 °C to 100 °C.
- 14 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 56 (4) , 1241-1253
A 0.5-V 560-kHz 18.8-fJ/Cycle On-Chip Oscillator With 96.1-ppm/°C Steady-State Stability Using a Duty-Cycled Digital Frequency-Locked Loop
On-chip oscillators are popular clocking solutions for a wide range of circuits and systems due to their ease of integration and low form factor, but their energy efficiency is typically limited to the pJ/cycle range by a number of contributors, such as active biasing currents, frequency dividers, and comparators. This work presents an on-chip oscillator for energy-efficient Internet-of-Things (IoT) applications based on a duty-cycled digital frequency-locked loop (DFLL) that reduces energy by disabling energy-hungry components and only periodically reactivating them to keep the output frequency stabilized during temperature drifts. A test chip is implemented in 65-nm CMOS and achieves 18.8 fJ/cycle (10.5 nW at 560 kHz) while maintaining an average steady-state temperature stability of 96.1 ppm/°C from 0 °C to 100 °C.
- 5 November 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 2 (12) , 301-304
A Piezoelectric Energy-Harvesting System With Parallel-SSHI Rectifier and Integrated Maximum-Power-Point Tracking
A piezoelectric energy-harvesting system, including a parallel-synchronized-switch harvesting-on-inductor (SSHI) rectifier and integrated maximum-power-point tracking (MPPT), is presented. The perturb & observe (P&O) algorithm is adopted for the MPPT of the parallel-SSHI rectifier. Furthermore, an output power evaluation algorithm for the P&O MPPT is proposed and its detailed implementation in the full analog domain is analyzed. Fabricated in 130-nm CMOS, measurement results show the piezoelectric harvesting system achieves a 417% energy-extraction improvement for the rectifier and a maximum 97% tracking efficiency for the MPPT, which makes the system achieve high energy-extraction improvement and MPPT efficiency simultaneously.
- 5 November 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 2 (12) , 301-304
A Piezoelectric Energy-Harvesting System With Parallel-SSHI Rectifier and Integrated Maximum-Power-Point Tracking
A piezoelectric energy-harvesting system, including a parallel-synchronized-switch harvesting-on-inductor (SSHI) rectifier and integrated maximum-power-point tracking (MPPT), is presented. The perturb & observe (P&O) algorithm is adopted for the MPPT of the parallel-SSHI rectifier. Furthermore, an output power evaluation algorithm for the P&O MPPT is proposed and its detailed implementation in the full analog domain is analyzed. Fabricated in 130-nm CMOS, measurement results show the piezoelectric harvesting system achieves a 417% energy-extraction improvement for the rectifier and a maximum 97% tracking efficiency for the MPPT, which makes the system achieve high energy-extraction improvement and MPPT efficiency simultaneously.
- Open Access
- 24 January 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Circuits and Systems
- Vol. 13 (2) , 271-281
A 1.02 μW Battery-Less, Continuous Sensing and Post-Processing SiP for Wearable Applications
Improving system lifetime and robustness is a key to advancing self-powered platforms for real world applications. A complete self-powered, battery-less, wearable platform requires a microwatt-power system-on-chip (SoC), operating reliably within this budget, capable of surviving long periods without charging, and recovering from power loss to its previous state. To meet these requirements, we designed a wireless sensing heterogeneous system-in-package (SiP) containing an ultra-low power (ULP) SoC, a non-volatile boot memory (NVM), and a 2.4 GHz frequency shift key (FSK) radio, all integrated with custom ULP interfaces. The SoC includes a fully integrated energy harvesting platform power manager (EH-PPM) to power the SiP and other commercial sensors. The EH-PPM is designed for small loads and powers the SoC and peripherals while drawing very low operating current. The SoC also includes a digital system data-flow for sensing applications, an analog front end for ECG signal acquisition, and a cold-boot management system (CBMS) for boot and recovery from the NVM. The CBMS enables integration of the SoC with the ULP NVM to create a wearable formfactor, self-powered system capable of recovery from power loss. The SoC also includes a radio interface tightly integrated with a compression accelerator to efficiently communicate with the FSK transmitter and reduce the FSK's transmission time. This tight integration between accelerators on the SoC and peripherals is another feature that reduces the system's power consumption by reducing the code size and number of memory accesses required to perform an operation. The SoC consumes 507 nW average power while running free-fall detection, 519 nW average power while measuring ambient temperature, and 1.02 μW during continuous ECG monitoring and post-processing.
- 24 January 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Circuits and Systems
- Vol. 13 (2) , 271-281
Open AccessA 1.02 μW Battery-Less, Continuous Sensing and Post-Processing SiP for Wearable Applications
Improving system lifetime and robustness is a key to advancing self-powered platforms for real world applications. A complete self-powered, battery-less, wearable platform requires a microwatt-power system-on-chip (SoC), operating reliably within this budget, capable of surviving long periods without charging, and recovering from power loss to its previous state. To meet these requirements, we designed a wireless sensing heterogeneous system-in-package (SiP) containing an ultra-low power (ULP) SoC, a non-volatile boot memory (NVM), and a 2.4 GHz frequency shift key (FSK) radio, all integrated with custom ULP interfaces. The SoC includes a fully integrated energy harvesting platform power manager (EH-PPM) to power the SiP and other commercial sensors. The EH-PPM is designed for small loads and powers the SoC and peripherals while drawing very low operating current. The SoC also includes a digital system data-flow for sensing applications, an analog front end for ECG signal acquisition, and a cold-boot management system (CBMS) for boot and recovery from the NVM. The CBMS enables integration of the SoC with the ULP NVM to create a wearable formfactor, self-powered system capable of recovery from power loss. The SoC also includes a radio interface tightly integrated with a compression accelerator to efficiently communicate with the FSK transmitter and reduce the FSK's transmission time. This tight integration between accelerators on the SoC and peripherals is another feature that reduces the system's power consumption by reducing the code size and number of memory accesses required to perform an operation. The SoC consumes 507 nW average power while running free-fall detection, 519 nW average power while measuring ambient temperature, and 1.02 μW during continuous ECG monitoring and post-processing.