Lennart Bamberg
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lennart Bamberg has a total of 50 co-authors. Here are the top co-authors by number of shared publications:
- Alberto Garcia-Ortiz
ITEM, University of Bremen, Bremen, Germany
22 shared publications - Jan Moritz Joseph
RWTH Aachen University Germany
9 shared publications - Thilo Pionteck
Institute for Information Technology and Communications, Otto-von-Guericke University Magdeburg, Magdeburg, Germany
9 shared publications - Amir Najafi
Univ Bremen, Inst Electrodynam & Microelect ITEM Ids, Otto Hahn Allee NW 1, D-28359 Bremen, Germany
7 shared publications - Ardalan Najafi
Institute of Electrodynamics and Microelectronics, University of Bremen, 28359 Bremen, Germany
6 shared publications Alberto Garcia-Ortiz
4 shared publications
Latest Publications
- Open Access
Rescaling-Aware Training for Efficient Deployment of Deep Learning Models on Full-Integer Hardware
Integer AI inference significantly reduces computational complexity in embedded systems. Quantization-aware training (QAT) helps mitigate accuracy degradation associated with post-training quantization but still overlooks the impact of integer rescaling during inference, which is a hardware costly operation in integer-only AI inference. This work shows that rescaling cost can be dramatically reduced post-training, by applying a stronger quantization to the rescale multiplicands at no model-quality loss. Furthermore, we introduce Rescale-Aware Training, a fine tuning method for ultra-low bit-width rescaling multiplicands. Experiments show that even with 8x reduced rescaler widths, the full accuracy is preserved through minimal incremental retraining. This enables more energy-efficient and cost-efficient AI inference for resource-constrained embedded systems.
Open AccessRescaling-Aware Training for Efficient Deployment of Deep Learning Models on Full-Integer Hardware
Integer AI inference significantly reduces computational complexity in embedded systems. Quantization-aware training (QAT) helps mitigate accuracy degradation associated with post-training quantization but still overlooks the impact of integer rescaling during inference, which is a hardware costly operation in integer-only AI inference. This work shows that rescaling cost can be dramatically reduced post-training, by applying a stronger quantization to the rescale multiplicands at no model-quality loss. Furthermore, we introduce Rescale-Aware Training, a fine tuning method for ultra-low bit-width rescaling multiplicands. Experiments show that even with 8x reduced rescaler widths, the full accuracy is preserved through minimal incremental retraining. This enables more energy-efficient and cost-efficient AI inference for resource-constrained embedded systems.
- Open Access
- 12 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Circuits and Systems
- Vol. 5, 178-188
Exploiting Neural-Network Statistics for Low-Power DNN Inference
Specialized compute blocks have been developed for efficient DNN execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power technique for edge-AI inference engines that combines overhead-free coding with a statistical analysis of the data and parameters of neural networks. Our approach reduces the power consumption of the logic, interconnect, and memory blocks used for data storage and movements by up to 80% for state-of-the-art benchmarks while providing additional power savings for the compute blocks by up to 39%. These power improvements are achieved with no loss of accuracy and negligible hardware cost.
- 12 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Circuits and Systems
- Vol. 5, 178-188
Open AccessExploiting Neural-Network Statistics for Low-Power DNN Inference
Specialized compute blocks have been developed for efficient DNN execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power technique for edge-AI inference engines that combines overhead-free coding with a statistical analysis of the data and parameters of neural networks. Our approach reduces the power consumption of the logic, interconnect, and memory blocks used for data storage and movements by up to 80% for state-of-the-art benchmarks while providing additional power savings for the compute blocks by up to 39%. These power improvements are achieved with no loss of accuracy and negligible hardware cost.
- 24 January 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 34 (4) , 1227-1240
Synapse Compression for Event-Based Convolutional-Neural-Network Accelerators
Manufacturing-viable neuromorphic chips require novel compute architectures to achieve the massively parallel and efficient information processing the brain supports so effortlessly. The most promising architectures for that are spiking/event-based, which enables massive parallelism at low complexity. However, the large memory requirements for synaptic connectivity are a showstopper for the execution of modern convolutional neural networks (CNNs) on massively parallel, event-based architectures. The present work overcomes this roadblock by contributing a lightweight hardware scheme to compress the synaptic memory requirements by several thousand times—enabling the execution of complex CNNs on a single chip of small form factor. A silicon implementation in a 12-nm technology shows that the technique achieves a total memory-footprint reduction of up to 374× compared to the best previously published technique at a negligible area overhead.
- 24 January 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 34 (4) , 1227-1240
Synapse Compression for Event-Based Convolutional-Neural-Network Accelerators
Manufacturing-viable neuromorphic chips require novel compute architectures to achieve the massively parallel and efficient information processing the brain supports so effortlessly. The most promising architectures for that are spiking/event-based, which enables massive parallelism at low complexity. However, the large memory requirements for synaptic connectivity are a showstopper for the execution of modern convolutional neural networks (CNNs) on massively parallel, event-based architectures. The present work overcomes this roadblock by contributing a lightweight hardware scheme to compress the synaptic memory requirements by several thousand times—enabling the execution of complex CNNs on a single chip of small form factor. A silicon implementation in a 12-nm technology shows that the technique achieves a total memory-footprint reduction of up to 374× compared to the best previously published technique at a negligible area overhead.
- Open Access
- 27 September 2021
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Modeling and Computer Simulation
- Vol. 32 (1) , 1-21
Ratatoskr: An Open-Source Framework for In-Depth Power, Performance, and Area Analysis and Optimization in 3D NoCs
We introduce Ratatoskr , an open-source framework for in-depth power, performance, and area (PPA) analysis in Networks-on-Chips (NoCs) for 3D-integrated and heterogeneous System-on-Chips (SoCs). It covers all layers of abstraction by providing an NoC hardware implementation on Register Transfer Level (RTL), an NoC simulator on cycle-accurate level and an application model on transaction level. By this comprehensive approach, Ratatoskr can provide the following specific PPA analyses: Dynamic power of links can be measured within 2.4% accuracy of bit-level simulations while maintaining cycle-accurate simulation speed. Router power is determined from RTL-to-gate-level synthesis combined with cycle-accurate simulations. The performance of the whole NoC can be measured both via cycle-accurate and RTL simulations. The performance (i.e., timing) of individual routers and the NoC area are obtained from RTL synthesis results. Despite these manifold features, Ratatoskr offers easy two-step user interaction: (1) A single point-of-entry allows setting design parameters. (2) PPA reports are generated automatically. For both the input and the output, different levels of abstraction can be chosen for high-level rapid network analysis or low-level improvement of architectural details. The synthesizable NoC-RTL model shows improved total router power and area in comparison to a conventional standard router. As a forward-thinking and unique feature not found in other NoC PPA-measurement tools, Ratatoskr supports heterogeneous 3D integration that is one of the most promising integration paradigms for upcoming SoCs. Thereby, Ratatoskr lays the groundwork to design their communication architectures. The framework is publicly available at https://github.com/ratatoskr-project .
- 27 September 2021
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Modeling and Computer Simulation
- Vol. 32 (1) , 1-21
Open AccessRatatoskr: An Open-Source Framework for In-Depth Power, Performance, and Area Analysis and Optimization in 3D NoCs
We introduce Ratatoskr , an open-source framework for in-depth power, performance, and area (PPA) analysis in Networks-on-Chips (NoCs) for 3D-integrated and heterogeneous System-on-Chips (SoCs). It covers all layers of abstraction by providing an NoC hardware implementation on Register Transfer Level (RTL), an NoC simulator on cycle-accurate level and an application model on transaction level. By this comprehensive approach, Ratatoskr can provide the following specific PPA analyses: Dynamic power of links can be measured within 2.4% accuracy of bit-level simulations while maintaining cycle-accurate simulation speed. Router power is determined from RTL-to-gate-level synthesis combined with cycle-accurate simulations. The performance of the whole NoC can be measured both via cycle-accurate and RTL simulations. The performance (i.e., timing) of individual routers and the NoC area are obtained from RTL synthesis results. Despite these manifold features, Ratatoskr offers easy two-step user interaction: (1) A single point-of-entry allows setting design parameters. (2) PPA reports are generated automatically. For both the input and the output, different levels of abstraction can be chosen for high-level rapid network analysis or low-level improvement of architectural details. The synthesizable NoC-RTL model shows improved total router power and area in comparison to a conventional standard router. As a forward-thinking and unique feature not found in other NoC PPA-measurement tools, Ratatoskr supports heterogeneous 3D integration that is one of the most promising integration paradigms for upcoming SoCs. Thereby, Ratatoskr lays the groundwork to design their communication architectures. The framework is publicly available at https://github.com/ratatoskr-project .
- 30 April 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 29 (6) , 1152-1163
High-Performance Logic-on-Memory Monolithic 3-D IC Designs for Arm Cortex-A Processors
Monolithic 3-D IC (M3-D) is a promising solution to improve the performance and energy-efficiency of modern processors. But, designers are faced with challenges in design tools and methodologies, especially for power and thermal verifications. We developed a new physical design flow that optimally places and routes cache modules in one tier and logic gates in the other. Our tool also builds high-quality clock and power delivery networks targeting logic-on-memory M3-D designs. Finally, we developed a sign-off analysis tool flow to evaluate power, performance, area (PPA), thermal, and voltage-drop quality for given M3-D designs. Using our complete register transfer level (RTL)-to-Graphic Design System (GDS) tool flow, we designed commercial quality 2-D and M3-D implementation of Arm Cortex-A7 and Cortex-A53 processors in a commercial 28-nm technology. Experimental results show that our 3-D processors offer 20% (A7) and 21% (A53) performance gain, compared with their 2-D commercial counterparts. The voltage-drop degradation of our 3-D Cortex-A7 and Cortex-A53 processors is less than 3% of the supply voltage, while temperature increase is 10.71 °C and 13.04 °C, respectively.
- 30 April 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 29 (6) , 1152-1163
High-Performance Logic-on-Memory Monolithic 3-D IC Designs for Arm Cortex-A Processors
Monolithic 3-D IC (M3-D) is a promising solution to improve the performance and energy-efficiency of modern processors. But, designers are faced with challenges in design tools and methodologies, especially for power and thermal verifications. We developed a new physical design flow that optimally places and routes cache modules in one tier and logic gates in the other. Our tool also builds high-quality clock and power delivery networks targeting logic-on-memory M3-D designs. Finally, we developed a sign-off analysis tool flow to evaluate power, performance, area (PPA), thermal, and voltage-drop quality for given M3-D designs. Using our complete register transfer level (RTL)-to-Graphic Design System (GDS) tool flow, we designed commercial quality 2-D and M3-D implementation of Arm Cortex-A7 and Cortex-A53 processors in a commercial 28-nm technology. Experimental results show that our 3-D processors offer 20% (A7) and 21% (A53) performance gain, compared with their 2-D commercial counterparts. The voltage-drop degradation of our 3-D Cortex-A7 and Cortex-A53 processors is less than 3% of the supply voltage, while temperature increase is 10.71 °C and 13.04 °C, respectively.
- 1 May 2020
- Published by Elsevier in Integration
- Vol. 72, 58-65
Misalignment-aware energy modeling of narrow buses for data encoding schemes
- 1 May 2020
- Published by Elsevier in Integration
- Vol. 72, 58-65
Misalignment-aware energy modeling of narrow buses for data encoding schemes
- Open Access
- 23 January 2020
- Published by MDPI AG in Technologies
- Vol. 8 (1) , 10
Application-Specific SoC Design Using Core Mapping to 3D Mesh NoCs with Nonlinear Area Optimization and Simulated Annealing
Core mapping, in which a core graph is mapped to a network graph to minimize communication, is a common design problem for Systems-on-Chip interconnected by a Network-on-Chip. In conventional multiprocessors, this mapping is area-agnostic as the cores in the core graph are uniform and therefore iso-area. This changes for Systems-on-Chip because tasks are mapped to specific blocks and not general-purpose cores. Thus, the area of these specific cores is varying. This requires novel mapping methods. In this paper, we propose a an area-aware cost function for simulated annealing; Furthermore, we advocate the use of nonlinear models as the area is nonlinear: A semi-definite program (SDP) can be used as it is sufficiently fast and shows 20% better area than conventional linear models. Our cost function allows for up to 16.4% better area, 2% better communication (bandwidth times hop distance) and 13.8% better total bandwidth in the network in comparison to the standard approach that accounts for both the network communication and uses cores with varying areas as well.
- 23 January 2020
- Published by MDPI AG in Technologies
- Vol. 8 (1) , 10
Open AccessApplication-Specific SoC Design Using Core Mapping to 3D Mesh NoCs with Nonlinear Area Optimization and Simulated Annealing
Core mapping, in which a core graph is mapped to a network graph to minimize communication, is a common design problem for Systems-on-Chip interconnected by a Network-on-Chip. In conventional multiprocessors, this mapping is area-agnostic as the cores in the core graph are uniform and therefore iso-area. This changes for Systems-on-Chip because tasks are mapped to specific blocks and not general-purpose cores. Thus, the area of these specific cores is varying. This requires novel mapping methods. In this paper, we propose a an area-aware cost function for simulated annealing; Furthermore, we advocate the use of nonlinear models as the area is nonlinear: A semi-definite program (SDP) can be used as it is sufficiently fast and shows 20% better area than conventional linear models. Our cost function allows for up to 16.4% better area, 2% better communication (bandwidth times hop distance) and 13.8% better total bandwidth in the network in comparison to the standard approach that accounts for both the network communication and uses cores with varying areas as well.
- 1 January 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 19 (1) , 51-54
Heterogeneous 3D Integration for a RISC-V System With STT-MRAM
Spin Torque Transfer Magnetic RAM (STT-MRAM) is a promising Non-Volatile Memory (NVM) technology achieving high density, low leakage power, and relatively small read/write delays. It provides a solution to improve the performance and to mitigate the leakage power consumption in SRAM-based processors. However, the process heterogeneity and the sophisticated back-end-of-line (BEOL) structure make it challenging to integrate the STT-MRAM in two-dimensional integrated circuits (2D ICs). In this paper, we implement a RISC-V-based system with STT-MRAM using a heterogeneous 3D integration methodology. Compared with the SRAM-based 2D counterpart, the MRAM-based 3D IC provides up to 17.55% silicon area saving, together with either 34.74% performance gain or 13.90% energy reduction.
- 1 January 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 19 (1) , 51-54
Heterogeneous 3D Integration for a RISC-V System With STT-MRAM
Spin Torque Transfer Magnetic RAM (STT-MRAM) is a promising Non-Volatile Memory (NVM) technology achieving high density, low leakage power, and relatively small read/write delays. It provides a solution to improve the performance and to mitigate the leakage power consumption in SRAM-based processors. However, the process heterogeneity and the sophisticated back-end-of-line (BEOL) structure make it challenging to integrate the STT-MRAM in two-dimensional integrated circuits (2D ICs). In this paper, we implement a RISC-V-based system with STT-MRAM using a heterogeneous 3D integration methodology. Compared with the SRAM-based 2D counterpart, the MRAM-based 3D IC provides up to 17.55% silicon area saving, together with either 34.74% performance gain or 13.90% energy reduction.