Anand Savanth
Anand Savanth has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Alex S. Weddell
University of Southampton
6 shared publications James Myers
Arm Research, Arm Ltd., Cambridge, U.K.
4 shared publications- Pranay Prabhat
Arm Research, Arm Ltd., Cambridge, U.K.
3 shared publications - Geoff V. Merrett
University of Southampton
2 shared publications Bashir M. Al-Hashimi
Center for IoT and Pervasive Systems, School of Electronics and Computer Science, University of Southampton, Southampton SO17 1BJ, U.K.
1 shared publication- Bashir M. Al-Hashimi
Electronics * CS, University of Southampton, Southampton, Hampshire United Kingdom of Great Britain and Northern Ireland
1 shared publication
Latest Publications
- 19 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 42 (1) , 95-108
Pragmatic Memory-System Support for Intermittent Computing Using Emerging Nonvolatile Memory
Intermittent computing (IC) is a key enabler for the vision of a trillion Internet of Things devices. By harvesting energy from the environment, and leveraging non-volatile memory (NVM) to retain computational progress across power cycles, IC enables untethered and battery-free devices to perform computation whenever ambient energy is available. The backbone of state retention is NVM, and recent advances in energy-efficient NVM have the potential to expand the application domain of IC significantly. Utilizing emerging NVM at the level of bitcells, researchers have proposed non-volatile processors. However, these do not leverage hardware-software co-design, which can be used to overcome hardware limitations and to provide support for application-level constraints such as atomicity. In this paper, we propose MEMIC, a memory architecture tailored for IC devices with byte-addressable NVM. A core focus of MEMIC is to combine volatile-and non-volatile memory in such a way that the operations of IC are as efficient as possible, while also maximizing computational performance per joule. MEMIC uses volatile memory for energy efficiency, and nonvolatile memory for data retention. To avoid double-buffered checkpoints and costly roll-backs when code needs to be reexecuted, MEMIC is designed to track and minimize writes to non-volatile memory during failure-atomic sections. Our evaluation shows that MEMIC’s instruction cache reduces workload completion time under intermittent operation by 41-70% and its data cache provides a further reduction of 13-39%.
- 19 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 42 (1) , 95-108
Pragmatic Memory-System Support for Intermittent Computing Using Emerging Nonvolatile Memory
Intermittent computing (IC) is a key enabler for the vision of a trillion Internet of Things devices. By harvesting energy from the environment, and leveraging non-volatile memory (NVM) to retain computational progress across power cycles, IC enables untethered and battery-free devices to perform computation whenever ambient energy is available. The backbone of state retention is NVM, and recent advances in energy-efficient NVM have the potential to expand the application domain of IC significantly. Utilizing emerging NVM at the level of bitcells, researchers have proposed non-volatile processors. However, these do not leverage hardware-software co-design, which can be used to overcome hardware limitations and to provide support for application-level constraints such as atomicity. In this paper, we propose MEMIC, a memory architecture tailored for IC devices with byte-addressable NVM. A core focus of MEMIC is to combine volatile-and non-volatile memory in such a way that the operations of IC are as efficient as possible, while also maximizing computational performance per joule. MEMIC uses volatile memory for energy efficiency, and nonvolatile memory for data retention. To avoid double-buffered checkpoints and costly roll-backs when code needs to be reexecuted, MEMIC is designed to track and minimize writes to non-volatile memory during failure-atomic sections. Our evaluation shows that MEMIC’s instruction cache reduces workload completion time under intermittent operation by 41-70% and its data cache provides a further reduction of 13-39%.
- 4 June 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 40 (3) , 444-452
Improving the Forward Progress of Transient Systems
Emerging applications for Internet of Things (IoT) devices demand smaller mass, size, and cost whilst increasing capability and reliability. Energy harvesting can provide power to these ultra-constrained devices, but introduces unreliability, unpredictability, and intermittency. Schemes for wireless sensors without batteries or supercapacitors overcome intermittency through saving system state into nonvolatile memory before the supply drops below the minimum operating voltage, termed transient, or intermittent computing. However, this introduces significant time and energy overheads. This article presents two schemes that significantly reduce these overheads: entering a sleep mode to avoid saving state and utilizing direct memory access (DMA) when state saves are required. Time and energy previously wasted on state saves can instead be used to perform useful computation, termed “forward progress.” We practically validate the proposed approaches across a range of energy sources and IoT benchmarks and demonstrate up to 46.8% and 40.3% increase in forward progress and up to 91.1% and 85.6% reduction in overheads for each scheme, respectively.
- 4 June 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 40 (3) , 444-452
Improving the Forward Progress of Transient Systems
Emerging applications for Internet of Things (IoT) devices demand smaller mass, size, and cost whilst increasing capability and reliability. Energy harvesting can provide power to these ultra-constrained devices, but introduces unreliability, unpredictability, and intermittency. Schemes for wireless sensors without batteries or supercapacitors overcome intermittency through saving system state into nonvolatile memory before the supply drops below the minimum operating voltage, termed transient, or intermittent computing. However, this introduces significant time and energy overheads. This article presents two schemes that significantly reduce these overheads: entering a sleep mode to avoid saving state and utilizing direct memory access (DMA) when state saves are required. Time and energy previously wasted on state saves can instead be used to perform useful computation, termed “forward progress.” We practically validate the proposed approaches across a range of energy sources and IoT benchmarks and demonstrate up to 46.8% and 40.3% increase in forward progress and up to 91.1% and 85.6% reduction in overheads for each scheme, respectively.
- 21 October 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 2 (9) , 175-178
A 0.98-nW/kHz 33-kHz Fully Integrated Subthreshold-Region Operation RC Oscillator With Forward-Body-Biasing
The lower bound on the power expended by an RC relaxation oscillator is decided by the RC network. This can be minimized by reducing the oscillation swing and increasing R. In the former technique, tighter comparator constraints limit power benefits while the latter technique increases resistor thermal noise bounding long-term jitter. To this end, this letter presents a fully integrated RC oscillator with core voltage aggressively scaled to subthreshold levels. A self-clocked switched-capacitor network is used to minimize voltage drop-out power loss. Full forward-body-biasing technique helps reduce device on-resistance. Additionally, temperature coefficient compensation for time constant is accomplished by poly resistors and a V TH -tracking reference scheme which avoids the use of diffusion resistors. This design is silicon-proven on 65-nm CMOS (0.0356-mm 2 area). The implementation has a 33-kHz clock with 32.2 nW at 1.2 V. Line sensitivity is within +0.7/−0.6% per volt across 16 samples for 1 to 1.5 V. Temperature sensitivity was measured to be 56 ppm/°C from 0 °C to 85 °C and measured Allan deviation < 100 ppm for averaging interval of $\tau = 400$ s and < 40 ppm for $\tau = 3000$ s.
- 21 October 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 2 (9) , 175-178
A 0.98-nW/kHz 33-kHz Fully Integrated Subthreshold-Region Operation RC Oscillator With Forward-Body-Biasing
The lower bound on the power expended by an RC relaxation oscillator is decided by the RC network. This can be minimized by reducing the oscillation swing and increasing R. In the former technique, tighter comparator constraints limit power benefits while the latter technique increases resistor thermal noise bounding long-term jitter. To this end, this letter presents a fully integrated RC oscillator with core voltage aggressively scaled to subthreshold levels. A self-clocked switched-capacitor network is used to minimize voltage drop-out power loss. Full forward-body-biasing technique helps reduce device on-resistance. Additionally, temperature coefficient compensation for time constant is accomplished by poly resistors and a V TH -tracking reference scheme which avoids the use of diffusion resistors. This design is silicon-proven on 65-nm CMOS (0.0356-mm 2 area). The implementation has a 33-kHz clock with 32.2 nW at 1.2 V. Line sensitivity is within +0.7/−0.6% per volt across 16 samples for 1 to 1.5 V. Temperature sensitivity was measured to be 56 ppm/°C from 0 °C to 85 °C and measured Allan deviation < 100 ppm for averaging interval of $\tau = 400$ s and < 40 ppm for $\tau = 3000$ s.
- 13 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 54 (11) , 3097-3106
A Sub-nW/kHz Relaxation Oscillator With Ratioed Reference and Sub-Clock Power Gated Comparator
Realizing the vision of a trillion IoT sensor nodes demands ultra low-power (ULP) compute, typically implemented using synchronous digital systems. These require low-power clock sources which must be fully integrated to meet low system-cost requirements. Hence, relaxation oscillators (RxOs) are popular. Their need for precision references and high-speed comparators can challenge power budgets; hence, prior works have explored low-speed closed-loop control for RxOs to alleviate this issue. This paper adopts an alternative approach to minimize the RxO power: the use of a ratioed switched-capacitor reference and sub-clock power-gating of the high-power comparator. The stability of the proposed design is further enhanced through the integration of digital-assist. This mixed-signal approach allowed a 1.2-MHz RxO to be implemented in 0.005 mm² in TSMC 65 nm, with silicon measurements showing 0.7%/V line and 100 ppm/°C temperature stability. The energy-per-cycle figure of merit is 0.68 nW/kHz--a 4x,improvement over state-of-the-art for comparable stability.
- 13 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 54 (11) , 3097-3106
A Sub-nW/kHz Relaxation Oscillator With Ratioed Reference and Sub-Clock Power Gated Comparator
Realizing the vision of a trillion IoT sensor nodes demands ultra low-power (ULP) compute, typically implemented using synchronous digital systems. These require low-power clock sources which must be fully integrated to meet low system-cost requirements. Hence, relaxation oscillators (RxOs) are popular. Their need for precision references and high-speed comparators can challenge power budgets; hence, prior works have explored low-speed closed-loop control for RxOs to alleviate this issue. This paper adopts an alternative approach to minimize the RxO power: the use of a ratioed switched-capacitor reference and sub-clock power-gating of the high-power comparator. The stability of the proposed design is further enhanced through the integration of digital-assist. This mixed-signal approach allowed a 1.2-MHz RxO to be implemented in 0.005 mm² in TSMC 65 nm, with silicon measurements showing 0.7%/V line and 100 ppm/°C temperature stability. The energy-per-cycle figure of merit is 0.68 nW/kHz--a 4x,improvement over state-of-the-art for comparable stability.
- 26 October 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 54 (2) , 550-559
Ultra-Low Power 18-Transistor Fully Static Contention-Free Single-Phase Clocked Flip-Flop in 65-nm CMOS
Flip-flops (FFs) are essential building blocks of sequential digital circuits but typically occupy a substantial proportion of chip area and consume significant amounts of power. This paper proposes 18-transistor single-phase clocked (18TSPC), a new topology of fully static contention-free single-phase clocked (SPC) FF with only 18 transistors, the lowest number reported for this type. Implemented in 65-nm CMOS, it achieves 20% cell area reduction compared to the conventional transmission gate FF (TGFF). Simulation results show the proposed 18TSPC is two times more efficient than TGFF in the energy-delay space. To demonstrate EDA compatibility and circuit/system-level benefits, a shift register and an AES-128 encryption engine have been implemented. Chip experimental measurements at 0.6 V, 25 °C show that, compared to TGFF, the proposed 18TSPC achieves reductions of 68% and 73% in overall and clock dynamic power, respectively, and 27% lower leakage.
- 26 October 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 54 (2) , 550-559
Ultra-Low Power 18-Transistor Fully Static Contention-Free Single-Phase Clocked Flip-Flop in 65-nm CMOS
Flip-flops (FFs) are essential building blocks of sequential digital circuits but typically occupy a substantial proportion of chip area and consume significant amounts of power. This paper proposes 18-transistor single-phase clocked (18TSPC), a new topology of fully static contention-free single-phase clocked (SPC) FF with only 18 transistors, the lowest number reported for this type. Implemented in 65-nm CMOS, it achieves 20% cell area reduction compared to the conventional transmission gate FF (TGFF). Simulation results show the proposed 18TSPC is two times more efficient than TGFF in the energy-delay space. To demonstrate EDA compatibility and circuit/system-level benefits, a shift register and an AES-128 encryption engine have been implemented. Chip experimental measurements at 0.6 V, 25 °C show that, compared to TGFF, the proposed 18TSPC achieves reductions of 68% and 73% in overall and clock dynamic power, respectively, and 27% lower leakage.
- Open Access
- 26 July 2018
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 1052 (1) , 012069
Energy Neutral Sensor System With Micro-scale Photovoltaic and Thermoelectric Energy Harvesting
Minimizing power conversion losses is critical for energy neutral operation of micro-scale energy harvested sensor nodes. These small form-factor sensor nodes rely on miniature harvesters with low output voltages that must be boosted with large conversion ratios to recharge batteries or super-capacitors. Selective Direct Operation (SDO), a technique to selectively avoid power conversion and thereby eliminate conversion loss in energy harvested systems has been demonstrated as an effective technique for light harvesters. This paper extends SDO to thermoelectric generators (TEGs). SDO exploits the ultra-low circuit functional voltages, enabling sensor systems to effectively harvest energy from cm-scale TEGs which output few 10's of mW but at low output voltages (100's of mV). PV cell construction from prior-work, TEG characterization and field measurements are presented in this paper to demonstrate the effectiveness of SDO and co-designing energy harvesters, power conversion circuits and digital sub-systems.
- 26 July 2018
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 1052 (1) , 012069
Open AccessEnergy Neutral Sensor System With Micro-scale Photovoltaic and Thermoelectric Energy Harvesting
Minimizing power conversion losses is critical for energy neutral operation of micro-scale energy harvested sensor nodes. These small form-factor sensor nodes rely on miniature harvesters with low output voltages that must be boosted with large conversion ratios to recharge batteries or super-capacitors. Selective Direct Operation (SDO), a technique to selectively avoid power conversion and thereby eliminate conversion loss in energy harvested systems has been demonstrated as an effective technique for light harvesters. This paper extends SDO to thermoelectric generators (TEGs). SDO exploits the ultra-low circuit functional voltages, enabling sensor systems to effectively harvest energy from cm-scale TEGs which output few 10's of mW but at low output voltages (100's of mV). PV cell construction from prior-work, TEG characterization and field measurements are presented in this paper to demonstrate the effectiveness of SDO and co-designing energy harvesters, power conversion circuits and digital sub-systems.
- 14 June 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 64 (9) , 2370-2379
Integrated Reciprocal Conversion With Selective Direct Operation for Energy Harvesting Systems
Energy harvesting IoT systems aim for energy neutrality, i.e., harvesting at least as much energy as is needed. This, however, is complicated by variations in environmental energy and application demands. Conventional systems use separate power converters to interface between the harvester and the storage, and then to the CPU system. Reciprocal power conversion has recently been proposed to perform both roles, eliminating redundancy and minimizing losses. This paper proposes to enhance this topology with “selective direct operation,” which completely bypasses the converter when appropriate. The integrated system, with 82% bidirectional conversion efficiency, was validated in 65-nm CMOS with only the harvester, battery, and decoupling capacitors being off-chip. Optimized for operation with cm2 photo-voltaic cell and a 32-b sub-threshold processor, the scheme enables up to 16% otherwise wasted energy to be utilized to provide >30% additional compute cycles under realistic indoor lighting conditions. Measured results show 84% peak conversion efficiency and energy neutral execution of benchmark sensor software (ULPBench) with cold-start capability.
- 14 June 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 64 (9) , 2370-2379
Integrated Reciprocal Conversion With Selective Direct Operation for Energy Harvesting Systems
Energy harvesting IoT systems aim for energy neutrality, i.e., harvesting at least as much energy as is needed. This, however, is complicated by variations in environmental energy and application demands. Conventional systems use separate power converters to interface between the harvester and the storage, and then to the CPU system. Reciprocal power conversion has recently been proposed to perform both roles, eliminating redundancy and minimizing losses. This paper proposes to enhance this topology with “selective direct operation,” which completely bypasses the converter when appropriate. The integrated system, with 82% bidirectional conversion efficiency, was validated in 65-nm CMOS with only the harvester, battery, and decoupling capacitors being off-chip. Optimized for operation with cm2 photo-voltaic cell and a 32-b sub-threshold processor, the scheme enables up to 16% otherwise wasted energy to be utilized to provide >30% additional compute cycles under realistic indoor lighting conditions. Measured results show 84% peak conversion efficiency and energy neutral execution of benchmark sensor software (ULPBench) with cold-start capability.