Guilherme Paim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Guilherme Paim has a total of 86 co-authors. Here are the top co-authors by number of shared publications:
- Sergio Bampi
Graduate Program in Microelectronics (PGMICRO), Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil
23 shared publications - Eduardo Antonio Cesar da Costa
Graduate Program on Electronic Engineering and Computing, Catholic University of Pelotas (UCPel), Pelotas, Brazil
17 shared publications - Leandro Mateus Giacomini Rocha
Graduate Program on Microelectronics (PGMicro) - Federal University of Rio Grande do Sul (UFRGS), Porto Alegre - Brazil
10 shared publications - Brunno Alves de Abreu
Graduate Program in Microelectronics (PGMICRO), Institute of Informatics (INF), Universidade Federal do Rio Grande do Sul (UFRGS), Porto Alegre, Brazil
8 shared publications - Hussam Amrouch
Technical University of Munich
6 shared publications Sergio Bampi
PGMICRO at Federal University of Rio Grande do Sul (UFRGS), Porto Alegre, Brazil
6 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 1055-1061
OpenGeMM: A Highly-Efficient GeMM Accelerator Generator with Lightweight RISC-V Control and Tight Memory Coupling
Deep neural networks (DNNs) face significant challenges when deployed on resource-constrained extreme edge devices due to their computational and data-intensive nature. While standalone accelerators tailored for specific application scenarios suffer from inflexible control and limited programmability, generic hardware acceleration platforms coupled with RISC-V CPUs can enable high reusability and flexibility, yet typically at the expense of system-level efficiency and low utilization. To fill this gap, we propose OpenGeMM, an open-source acceleration platform, jointly demonstrating high efficiency and utilization, as well as ease of configurability and programmability. OpenGeMM encompasses a parameterized Chisel-coded GeMM accelerator, a lightweight RISC-V processor, and a tightly coupled multi-banked scratchpad memory. The GeMM core utilization and system efficiency are boosted through three mechanisms: configuration pre-loading, input pre-fetching with output buffering, and programmable strided memory access. Experimental results show that OpenGeMM can consistently achieve hardware utilization ranging from 81.89% to 99.34% across diverse CNN and Transformer workloads. Compared to the SotA open-source Gemmini accelerator, OpenGeMM demonstrates a 3.58× to 16.40× speedup on normalized throughput across a wide variety of GeMM workloads, while achieving 4.68 TOPS/W system efficiency.
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 1055-1061
Top Cited PapersOpen AccessOpenGeMM: A Highly-Efficient GeMM Accelerator Generator with Lightweight RISC-V Control and Tight Memory Coupling
Deep neural networks (DNNs) face significant challenges when deployed on resource-constrained extreme edge devices due to their computational and data-intensive nature. While standalone accelerators tailored for specific application scenarios suffer from inflexible control and limited programmability, generic hardware acceleration platforms coupled with RISC-V CPUs can enable high reusability and flexibility, yet typically at the expense of system-level efficiency and low utilization. To fill this gap, we propose OpenGeMM, an open-source acceleration platform, jointly demonstrating high efficiency and utilization, as well as ease of configurability and programmability. OpenGeMM encompasses a parameterized Chisel-coded GeMM accelerator, a lightweight RISC-V processor, and a tightly coupled multi-banked scratchpad memory. The GeMM core utilization and system efficiency are boosted through three mechanisms: configuration pre-loading, input pre-fetching with output buffering, and programmable strided memory access. Experimental results show that OpenGeMM can consistently achieve hardware utilization ranging from 81.89% to 99.34% across diverse CNN and Transformer workloads. Compared to the SotA open-source Gemmini accelerator, OpenGeMM demonstrates a 3.58× to 16.40× speedup on normalized throughput across a wide variety of GeMM workloads, while achieving 4.68 TOPS/W system efficiency.
- Open Access
- 11 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 10, 125-134
System-Technology Co-Optimization for Dense Edge Architectures Using 3-D Integration and Nonvolatile Memory
High-performance edge artificial intelligence (Edge-AI) inference applications aim for high energy efficiency, memory density and small form factor, requiring a design space exploration across the whole stack – workloads, architecture, mapping and co-optimization with emerging technology. In this paper, we present an system-technology co-optimization (STCO) framework that interfaces with workload-driven system scaling challenges and physical design-enabled technology offerings. The framework is built on three engines that provide the physical design characterization, dataflow mapping optimizer, and system efficiency predictor. The framework builds on a systolic array accelerator to provide the design-technology characterization points using advanced imec A10 nanosheet CMOS node along with emerging, high-density voltage-gated spin-orbit-torque (VGSOT) MRAM, combined with memory-on-logic fine-pitch 3D wafer-to-wafer hybrid bonding. We observe that 3D system integration of SRAM-based design leads to 9% power savings with 53% footprint reduction at iso-frequency w.r.t. 2D implementation for the same memory capacity. 3D NVM-VGSOT allows 4× memory capacity increase with 30% footprint reduction at iso-power compared to 2D SRAM 1×. Our exploration with two diverse workloads – image resolution enhancement (FSRCNN) and eye tracking (EDSNet) – shows that more resources allow better workload mapping possibilities which are able to compensate peak system energy efficiency degradation on high memory capacity cases. We show that a 25% peak efficiency reduction on a 32× memory capacity can lead to a 7.4× faster execution with 5.7× higher effective TOPS/W than the 1× memory capacity case on the same technology.
- 11 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 10, 125-134
Open AccessSystem-Technology Co-Optimization for Dense Edge Architectures Using 3-D Integration and Nonvolatile Memory
High-performance edge artificial intelligence (Edge-AI) inference applications aim for high energy efficiency, memory density and small form factor, requiring a design space exploration across the whole stack – workloads, architecture, mapping and co-optimization with emerging technology. In this paper, we present an system-technology co-optimization (STCO) framework that interfaces with workload-driven system scaling challenges and physical design-enabled technology offerings. The framework is built on three engines that provide the physical design characterization, dataflow mapping optimizer, and system efficiency predictor. The framework builds on a systolic array accelerator to provide the design-technology characterization points using advanced imec A10 nanosheet CMOS node along with emerging, high-density voltage-gated spin-orbit-torque (VGSOT) MRAM, combined with memory-on-logic fine-pitch 3D wafer-to-wafer hybrid bonding. We observe that 3D system integration of SRAM-based design leads to 9% power savings with 53% footprint reduction at iso-frequency w.r.t. 2D implementation for the same memory capacity. 3D NVM-VGSOT allows 4× memory capacity increase with 30% footprint reduction at iso-power compared to 2D SRAM 1×. Our exploration with two diverse workloads – image resolution enhancement (FSRCNN) and eye tracking (EDSNet) – shows that more resources allow better workload mapping possibilities which are able to compensate peak system energy efficiency degradation on high memory capacity cases. We show that a 25% peak efficiency reduction on a 32× memory capacity can lead to a 7.4× faster execution with 5.7× higher effective TOPS/W than the 1× memory capacity case on the same technology.
- 1 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (1) , 180-191
On the Efficacy and Vulnerabilities of Logic Locking in Tree-Based Machine Learning
The popularity and widespread usage of machine learning (ML) hardware have created challenges for its intellectual property (IP) protection. Logic locking is a widely used technique for IP protection but has received little attention in error-resilient applications such as ML hardware modules. This work investigates the effectiveness of logic locking when applied to tree-based ML circuits and reveals a critical vulnerability that undermines its effectiveness for single-label ML classifiers. We propose a logic locking scheme to eliminate the vulnerabilities in decision trees (DTs) and random forests (RFs) circuits. In our extensive simulation involving 16 DTs and 16 RFs, our solution consistently thwarts the vulnerability. We further evaluated the security of our approach by considering different obfuscation percentages and launching state-of-the-art oracle-less attacks on logic locking. Our method proves resilient, indicating that by fixing the identified vulnerability, we did not introduce new attack vectors. Further, our investigation indicates that DT/RF accelerators are significantly less vulnerable to oracle-less attacks compared to exact circuits. Overall, our work lays the foundation for future investigations into the effectiveness of logic locking for ML circuits.
- 1 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (1) , 180-191
On the Efficacy and Vulnerabilities of Logic Locking in Tree-Based Machine Learning
The popularity and widespread usage of machine learning (ML) hardware have created challenges for its intellectual property (IP) protection. Logic locking is a widely used technique for IP protection but has received little attention in error-resilient applications such as ML hardware modules. This work investigates the effectiveness of logic locking when applied to tree-based ML circuits and reveals a critical vulnerability that undermines its effectiveness for single-label ML classifiers. We propose a logic locking scheme to eliminate the vulnerabilities in decision trees (DTs) and random forests (RFs) circuits. In our extensive simulation involving 16 DTs and 16 RFs, our solution consistently thwarts the vulnerability. We further evaluated the security of our approach by considering different obfuscation percentages and launching state-of-the-art oracle-less attacks on logic locking. Our method proves resilient, indicating that by fixing the identified vulnerability, we did not introduce new attack vectors. Further, our investigation indicates that DT/RF accelerators are significantly less vulnerable to oracle-less attacks compared to exact circuits. Overall, our work lays the foundation for future investigations into the effectiveness of logic locking for ML circuits.
- 18 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Embedded Systems Letters
- Vol. 16 (3) , 279-282
Improving Netlist Transformation-Based Approximate Logic Synthesis Through Resynthesis
To address the challenges of efficient hardware design for error-tolerant applications, several techniques of applied approximate computing have been proposed. Pruning algorithms aim to approximate circuits with reduced design requirements at the cost of an acceptable degradation of their quality-of-result. In this letter, we present the effects of resynthesis, an iterative application of logic synthesis along with pruning algorithms, into a state-of-the-art approximate design flow, AxLS. Resynthesis strategy improve the approximation, achieving up to 70% area-power savings for the same error in the output, and reducing the number of iterations, and hence the time required to explore the design space in up to 30×, to obtain an approximated design.
- 18 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Embedded Systems Letters
- Vol. 16 (3) , 279-282
Improving Netlist Transformation-Based Approximate Logic Synthesis Through Resynthesis
To address the challenges of efficient hardware design for error-tolerant applications, several techniques of applied approximate computing have been proposed. Pruning algorithms aim to approximate circuits with reduced design requirements at the cost of an acceptable degradation of their quality-of-result. In this letter, we present the effects of resynthesis, an iterative application of logic synthesis along with pruning algorithms, into a state-of-the-art approximate design flow, AxLS. Resynthesis strategy improve the approximation, achieving up to 70% area-power savings for the same error in the output, and reducing the number of iterations, and hence the time required to explore the design space in up to 30×, to obtain an approximated design.
- 26 February 2024
- Published by Springer Nature in Circuits, Systems, and Signal Processing
- Vol. 43 (6) , 3649-3678
AxRSU-$2^{m}$: Higher-Order m-Bit Approximate Encoders for Radix-$2^{m}$ Squarer Units- 26 February 2024
- Published by Springer Nature in Circuits, Systems, and Signal Processing
- Vol. 43 (6) , 3649-3678
AxRSU-$2^{m}$: Higher-Order m-Bit Approximate Encoders for Radix-$2^{m}$ Squarer Units- 6 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 13 (1) , 86-95
Compact CMOS-Compatible Majority Gate Using Body Biasing in FDSOI Technology
This is the first work that employs the body biasing feature, available in fully depleted silicon on insulator (FDSOI) technology nodes, to build CMOS-compatible compact majority (MAJ) and minority (MIN) logic gates. Compared to their CMOS counterpart, our novel MAJ/MIN gates exhibit considerably fewer transistors, 2.3x and 2x, respectively. The use of MAJ/MIN gates for logic synthesis has been widely investigated in the literature since any function can be realized using only these two gates. Nevertheless, the use of MAJ/MIN gates to synthesize circuits remained very limited because such gates are area hungry when implemented using conventional standard cells. Therefore, state-of-the-art research in the last decade has extensively focused on exploring beyond-CMOS technologies to build alternative MAJ/MIN gates. However, such emerging technologies are still immature and suffering considerably from variability. On the other hand, our novel FDSOI-based MAJ/MIN gates are based on mature CMOS commercial technologies. Our analysis of the proposed gates has been conducted using accurate SPICE simulations with an industry-compact transistor model, BSIM-IMG, calibrated for an industrial 14nm FDSOI technology. Using our developed MAJ/MIN synthesis and error injection framework called MAJinBoo, we demonstrate that MAJ/MIN-based circuits exhibit an outstanding resiliency against errors due to the inherent voting-based computing. This encourages the employment of our MAJ/MIN gates in safety-critical applications where reliability is a prime concern. Our MAJinBoo framework is available at https://github.com/Brunno815/MAJinBoo.
- 6 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 13 (1) , 86-95
Compact CMOS-Compatible Majority Gate Using Body Biasing in FDSOI Technology
This is the first work that employs the body biasing feature, available in fully depleted silicon on insulator (FDSOI) technology nodes, to build CMOS-compatible compact majority (MAJ) and minority (MIN) logic gates. Compared to their CMOS counterpart, our novel MAJ/MIN gates exhibit considerably fewer transistors, 2.3x and 2x, respectively. The use of MAJ/MIN gates for logic synthesis has been widely investigated in the literature since any function can be realized using only these two gates. Nevertheless, the use of MAJ/MIN gates to synthesize circuits remained very limited because such gates are area hungry when implemented using conventional standard cells. Therefore, state-of-the-art research in the last decade has extensively focused on exploring beyond-CMOS technologies to build alternative MAJ/MIN gates. However, such emerging technologies are still immature and suffering considerably from variability. On the other hand, our novel FDSOI-based MAJ/MIN gates are based on mature CMOS commercial technologies. Our analysis of the proposed gates has been conducted using accurate SPICE simulations with an industry-compact transistor model, BSIM-IMG, calibrated for an industrial 14nm FDSOI technology. Using our developed MAJ/MIN synthesis and error injection framework called MAJinBoo, we demonstrate that MAJ/MIN-based circuits exhibit an outstanding resiliency against errors due to the inherent voting-based computing. This encourages the employment of our MAJ/MIN gates in safety-critical applications where reliability is a prime concern. Our MAJinBoo framework is available at https://github.com/Brunno815/MAJinBoo.
- 3 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 71 (11) , 4748-4752
MAxPy: A Framework for Bridging Approximate Computing Circuits to Its Applications
This brief presents MAxPy, a framework for bridging approximate computing (AxC) circuit design to its applications. MAxPy is an application-agnostic framework able to automatically build a cycle-accurate Python model of an approximate hardware design. This model can easily be emulated and integrated as a module in Python-based applications. We are herein proposing this framework aiming to build an AxC toolbox stimulating open research in academia to promote the integration of the existing and future open-source (i) AxC benchmarks, (ii) approximate logic synthesis tools, and (iii) approximate arithmetic unit insights. In this brief, we demonstrate the use of MAxPy to explore the AxC design space of a Sobel filter hardware design as a case study exploiting a set of approximate adders combined with the data-driven approximate logic synthesis via probabilistic pruning. Then, we present a Pareto front for circuit area, energy, and delay reduction versus the application-level metrics: edge detection accuracy and structural similarity index (SSIM). The Pareto front results of the multiple AxC techniques herein explored show circuit area savings ranging from 40.4% to 59.1% for a 98.1% to 99.6% accuracy on edge detection application. MAxPy code is open source in: github.com/MAxPy-Project.
- 3 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 71 (11) , 4748-4752
MAxPy: A Framework for Bridging Approximate Computing Circuits to Its Applications
This brief presents MAxPy, a framework for bridging approximate computing (AxC) circuit design to its applications. MAxPy is an application-agnostic framework able to automatically build a cycle-accurate Python model of an approximate hardware design. This model can easily be emulated and integrated as a module in Python-based applications. We are herein proposing this framework aiming to build an AxC toolbox stimulating open research in academia to promote the integration of the existing and future open-source (i) AxC benchmarks, (ii) approximate logic synthesis tools, and (iii) approximate arithmetic unit insights. In this brief, we demonstrate the use of MAxPy to explore the AxC design space of a Sobel filter hardware design as a case study exploiting a set of approximate adders combined with the data-driven approximate logic synthesis via probabilistic pruning. Then, we present a Pareto front for circuit area, energy, and delay reduction versus the application-level metrics: edge detection accuracy and structural similarity index (SSIM). The Pareto front results of the multiple AxC techniques herein explored show circuit area savings ranging from 40.4% to 59.1% for a 98.1% to 99.6% accuracy on edge detection application. MAxPy code is open source in: github.com/MAxPy-Project.
- 3 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 71 (11) , 4768-4772
VLSI Architecture for Energy-Efficient and Accurate Pre-Processing Pan-Tompkins Design
This brief proposes an energy-efficient and accurate VLSI architecture of the pre-processing Pan-Tompkins algorithm (PTA) for filtering electrocardiogram (ECG) signals. The PTA is among the most straightforward algorithms to detect the QRS complex in ECG signals. This brief proposes a fully-parallel PTA architecture with a reduced amount of registers, directly impacting energy reduction. Furthermore, we implement a unified band-pass filter exploring shift-add arrangement to reduce the noises in the ECG signal. This strategy contributes to the savings of 46.49% and 34.64% in the area and energy per operation (EPO), respectively, compared to the literature, keeping the average sensitivity and positive prediction for examined samples at 99.73% and 99.71%, respectively.
- 3 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 71 (11) , 4768-4772
VLSI Architecture for Energy-Efficient and Accurate Pre-Processing Pan-Tompkins Design
This brief proposes an energy-efficient and accurate VLSI architecture of the pre-processing Pan-Tompkins algorithm (PTA) for filtering electrocardiogram (ECG) signals. The PTA is among the most straightforward algorithms to detect the QRS complex in ECG signals. This brief proposes a fully-parallel PTA architecture with a reduced amount of registers, directly impacting energy reduction. Furthermore, we implement a unified band-pass filter exploring shift-add arrangement to reduce the noises in the ECG signal. This strategy contributes to the savings of 46.49% and 34.64% in the area and energy per operation (EPO), respectively, compared to the literature, keeping the average sensitivity and positive prediction for examined samples at 99.73% and 99.71%, respectively.
- 31 January 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 70 (3) , 1264-1268
Robustness Analysis of 3–2 Adder Compressor Designed in 7-nm FinFET Technology
Adder compressors (ACs) have been employed for several prominent computing datapaths as multipliers, machine learning, discrete transforms, and filters. This paper investigates the robustness of the 3-2 AC against the process, voltage, and temperature (PVT) variability targeting a design for a predictive ASAP7 7nm FinFET technology. The variability impacts on the delay, power, and product-delay-power (PDP) of the 3-2 AC for both super-and near-threshold voltages operating regimes are herein evaluated. Our results demonstrate that process is the main concern among the variability issues for both operating regimes, presenting an even more alarming level of variability in near-threshold operation.
- 31 January 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 70 (3) , 1264-1268
Robustness Analysis of 3–2 Adder Compressor Designed in 7-nm FinFET Technology
Adder compressors (ACs) have been employed for several prominent computing datapaths as multipliers, machine learning, discrete transforms, and filters. This paper investigates the robustness of the 3-2 AC against the process, voltage, and temperature (PVT) variability targeting a design for a predictive ASAP7 7nm FinFET technology. The variability impacts on the delay, power, and product-delay-power (PDP) of the 3-2 AC for both super-and near-threshold voltages operating regimes are herein evaluated. Our results demonstrate that process is the main concern among the variability issues for both operating regimes, presenting an even more alarming level of variability in near-threshold operation.
- 22 November 2022
- Published by Springer Nature in Circuits, Systems, and Signal Processing
- Vol. 42 (2) , 828-852
Energy-Efficient VLSI Squarer Unit with Optimized Radix-2m Multiplication Logic
- 22 November 2022
- Published by Springer Nature in Circuits, Systems, and Signal Processing
- Vol. 42 (2) , 828-852
Energy-Efficient VLSI Squarer Unit with Optimized Radix-2m Multiplication Logic