Amit Kulkarni
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
Research Subjects
Amit Kulkarni has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
Dirk Stroobandt
ELIS department, Computer Systems Lab, Ghent University, iGent, Technologiepark - Zwijnaarde 15, B-9052 Ghent - Belgium
4 shared publications- Dirk Stroobandt
Ghent University Belgium
3 shared publications Ana Lucia Varbanescu
Faculty of Science IVI, University of Amsterdam, Postbus 94323 1090 GH Amsterdam - The Netherlands
1 shared publicationAndre Werner
1 shared publication- Cătălin Bogdan Ciobanu
University of Amsterdam
1 shared publication Florian Fricke
1 shared publication
Latest Publications
- 1 November 2019
- Published by Elsevier in Journal of Parallel and Distributed Computing
- Vol. 133, 420-431
Reco-Pi: A reconfigurable Cryptoprocessor for -Cipher- 1 November 2019
- Published by Elsevier in Journal of Parallel and Distributed Computing
- Vol. 133, 420-431
Reco-Pi: A reconfigurable Cryptoprocessor for -Cipher- Open Access
- 9 September 2019
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Reconfigurable Technology and Systems
- Vol. 12 (4) , 1-27
Distributed Inference over Decision Tree Ensembles on Clusters of FPGAs
Given the growth in data inputs and application complexity, it is often the case that a single hardware accelerator is not enough to solve a given problem. In particular, the computational demands and I/O of many tasks in machine learning often require a cluster of accelerators to make a relevant difference in performance. In this article, we explore the efficient construction of FPGA clusters using inference over Decision Tree Ensembles as the target application. The article explores several levels of the problem: (1) a lightweight inter-FPGA communication protocol and routing layer to facilitate the communication between the different FPGAs, (2) the data partitioning and distribution strategies maximizing performance, (3) and an in depth analysis on how applications can be efficiently distributed over such a cluster. The experimental analysis shows that the resulting system can support inference over decision tree ensembles at a significantly higher throughput than that achieved by existing systems.
- 9 September 2019
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Reconfigurable Technology and Systems
- Vol. 12 (4) , 1-27
Open AccessDistributed Inference over Decision Tree Ensembles on Clusters of FPGAs
Given the growth in data inputs and application complexity, it is often the case that a single hardware accelerator is not enough to solve a given problem. In particular, the computational demands and I/O of many tasks in machine learning often require a cluster of accelerators to make a relevant difference in performance. In this article, we explore the efficient construction of FPGA clusters using inference over Decision Tree Ensembles as the target application. The article explores several levels of the problem: (1) a lightweight inter-FPGA communication protocol and routing layer to facilitate the communication between the different FPGAs, (2) the data partitioning and distribution strategies maximizing performance, (3) and an in depth analysis on how applications can be efficiently distributed over such a cluster. The experimental analysis shows that the resulting system can support inference over decision tree ensembles at a significantly higher throughput than that achieved by existing systems.
- 1 December 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 203-206
A NoC-based custom FPGA configuration memory architecture for ultra-fast micro-reconfiguration
Run-time reconfiguration in FPGAs is an important feature that offers design flexibility under low-cost silicon area and power budgets, at the cost of reconfiguration overhead. The reconfiguration time overhead produced by the conventional configuration ports (such as ICAP) is too high for the reconfiguration technology to be embraced as a standard. Furthermore, the current FPGA configuration memory architecture restricts the access of configuration data to the frame level; this significantly delays the reconfiguration process. The work presented in this paper explores the design space of the configuration memory architecture that fits the design of large FPGA's and is suitable to accomplish needs for ultra-fast reconfiguration. Therefore, the proposed method could be a stepping stone for next generation FPGA configuration memory architecture. Our simulation results show a reconfiguration speed gain of a factor of at least 1000 for substantially big parameterized applications that come with the cost of extra auxiliary hardware used on top of the column-based FPGA architecture.
- 1 December 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 203-206
A NoC-based custom FPGA configuration memory architecture for ultra-fast micro-reconfiguration
Run-time reconfiguration in FPGAs is an important feature that offers design flexibility under low-cost silicon area and power budgets, at the cost of reconfiguration overhead. The reconfiguration time overhead produced by the conventional configuration ports (such as ICAP) is too high for the reconfiguration technology to be embraced as a standard. Furthermore, the current FPGA configuration memory architecture restricts the access of configuration data to the frame level; this significantly delays the reconfiguration process. The work presented in this paper explores the design space of the configuration memory architecture that fits the design of large FPGA's and is suitable to accomplish needs for ultra-fast reconfiguration. Therefore, the proposed method could be a stepping stone for next generation FPGA configuration memory architecture. Our simulation results show a reconfiguration speed gain of a factor of at least 1000 for substantially big parameterized applications that come with the cost of extra auxiliary hardware used on top of the column-based FPGA architecture.
- Open Access
Pixie: A heterogeneous Virtual Coarse-Grained Reconfigurable Array for high performance image processing applications
Coarse-Grained Reconfigurable Arrays (CGRAs) enable ease of programmability and result in low development costs. They enable the ease of use specifically in reconfigurable computing applications. The smaller cost of compilation and reduced reconfiguration overhead enables them to become attractive platforms for accelerating high-performance computing applications such as image processing. The CGRAs are ASICs and therefore, expensive to produce. However, Field Programmable Gate Arrays (FPGAs) are relatively cheaper for low volume products but they are not so easily programmable. We combine best of both worlds by implementing a Virtual Coarse-Grained Reconfigurable Array (VCGRA) on FPGA. VCGRAs are a trade off between FPGA with large routing overheads and ASICs. In this perspective we present a novel heterogeneous Virtual Coarse-Grained Reconfigurable Array (VCGRA) called "Pixie" which is suitable for implementing high performance image processing applications. The proposed VCGRA contains generic processing elements and virtual channels that are described using the Hardware Description Language VHDL. Both elements have been optimized by using the parameterized configuration tool flow and result in a resource reduction of 24% for each processing elements and 82% for each virtual channels respectively.
Open AccessPixie: A heterogeneous Virtual Coarse-Grained Reconfigurable Array for high performance image processing applications
Coarse-Grained Reconfigurable Arrays (CGRAs) enable ease of programmability and result in low development costs. They enable the ease of use specifically in reconfigurable computing applications. The smaller cost of compilation and reduced reconfiguration overhead enables them to become attractive platforms for accelerating high-performance computing applications such as image processing. The CGRAs are ASICs and therefore, expensive to produce. However, Field Programmable Gate Arrays (FPGAs) are relatively cheaper for low volume products but they are not so easily programmable. We combine best of both worlds by implementing a Virtual Coarse-Grained Reconfigurable Array (VCGRA) on FPGA. VCGRAs are a trade off between FPGA with large routing overheads and ASICs. In this perspective we present a novel heterogeneous Virtual Coarse-Grained Reconfigurable Array (VCGRA) called "Pixie" which is suitable for implementing high performance image processing applications. The proposed VCGRA contains generic processing elements and virtual channels that are described using the Hardware Description Language VHDL. Both elements have been optimized by using the parameterized configuration tool flow and result in a resource reduction of 24% for each processing elements and 82% for each virtual channels respectively.
- 8 October 2016
- Published by Springer Nature in Design Automation for Embedded Systems
- Vol. 20 (4) , 341-359
MiCAP-Pro: a high speed custom reconfiguration controller for Dynamic Circuit Specialization
- 8 October 2016
- Published by Springer Nature in Design Automation for Embedded Systems
- Vol. 20 (4) , 341-359
MiCAP-Pro: a high speed custom reconfiguration controller for Dynamic Circuit Specialization
- Open Access
- 1 January 2016
- Published by Wiley in International Journal of Reconfigurable Computing
- Vol. 2016, 1-12
How to Efficiently Reconfigure Tunable Lookup Tables for Dynamic Circuit Specialization
Dynamic Circuit Specialization is used to optimize the implementation of a parameterized application on an FPGA. Instead of implementing the parameters as regular inputs, in the DCS approach these inputs are implemented as constants. When the parameter values change, the design is reoptimized for the new constant values by reconfiguring the FPGA. This allows faster and more resource-efficient implementation but investigations have shown that reconfiguration time is the major limitation for DCS implementation on Xilinx FPGAs. The limitation arises from the use of inefficient reconfiguration methods in conventional DCS implementation. To address this issue, we propose different approaches to reduce the reconfiguration time drastically and improve the reconfiguration speed. In this context, this paper presents the use of custom reconfiguration controllers and custom reconfiguration software drivers, along with placement constraints to shorten the reconfiguration time. Our results show an improvement in the reconfiguration speed by at least a factor 14 by using Xilinx reconfiguration controller along with placement constraints. However, the improvement can go up to a factor 40 with the combination of a custom reconfiguration controller, custom software drivers, and placement constraints. We also observe depreciation in the system’s performance by at least 6% due to placement constraints.
- 1 January 2016
- Published by Wiley in International Journal of Reconfigurable Computing
- Vol. 2016, 1-12
Open AccessHow to Efficiently Reconfigure Tunable Lookup Tables for Dynamic Circuit Specialization
Dynamic Circuit Specialization is used to optimize the implementation of a parameterized application on an FPGA. Instead of implementing the parameters as regular inputs, in the DCS approach these inputs are implemented as constants. When the parameter values change, the design is reoptimized for the new constant values by reconfiguring the FPGA. This allows faster and more resource-efficient implementation but investigations have shown that reconfiguration time is the major limitation for DCS implementation on Xilinx FPGAs. The limitation arises from the use of inefficient reconfiguration methods in conventional DCS implementation. To address this issue, we propose different approaches to reduce the reconfiguration time drastically and improve the reconfiguration speed. In this context, this paper presents the use of custom reconfiguration controllers and custom reconfiguration software drivers, along with placement constraints to shorten the reconfiguration time. Our results show an improvement in the reconfiguration speed by at least a factor 14 by using Xilinx reconfiguration controller along with placement constraints. However, the improvement can go up to a factor 40 with the combination of a custom reconfiguration controller, custom software drivers, and placement constraints. We also observe depreciation in the system’s performance by at least 6% due to placement constraints.
- 1 December 2015
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Power measurements and analysis for dynamic circuit specialization
Dynamic Circuit Specialization (DCS) is a technique for optimized FPGA implementation and is built on top of Partial Reconfiguration (PR). Dynamic Partial Reconfiguration (DPR) provides an opportunity to share the silicon area between different Partially Reconfigurable Modules (PRMs) and therefore results in smaller and faster designs that potentially also reduce the power consumption. In this paper, we show that energy consumption is an important factor that has to be considered while implementing a parameterized design using DCS. In order to make a good choice for implementing a parameterized design with the goal of power optimized implementation, it is important to have a good power consumption estimate of the Dynamic Circuit Specialization. In this context, our paper presents a detailed investigation of the power consumption of a parameterized design implemented using DCS on the Xilinx Zynq-SoC FPGA. We propose an energy analysis of DCS and investigate the benefits of the use of DCS in comparison with a classic static FPGA implementation. We see that the power needed for the reconfiguration is much higher than the gain in power using the reconfiguration over the static implementation. An important reason is because of the CPU involved during the reconfiguration and the interface between the AXI bus and the HWICAP. To reduce the reconfiguration power, we include a clock gating technique to the reconfiguration interface AXIHWICAP that makes DCS more power efficient. We also relate the power gain to the size of the implementation and to the allowed time to reconfigure versus the useful run time. We conclude that for an implementation with 10 FIR filters, the reconfiguration time should not take more than 30.3% of the total time in order to remain energy efficient. Considering a specific use case with 10 FIR filters at a reconfiguration rate of 0.01, the energy consumption using DCS implementation is 20.5% lower than using the static FIR.
- 1 December 2015
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Power measurements and analysis for dynamic circuit specialization
Dynamic Circuit Specialization (DCS) is a technique for optimized FPGA implementation and is built on top of Partial Reconfiguration (PR). Dynamic Partial Reconfiguration (DPR) provides an opportunity to share the silicon area between different Partially Reconfigurable Modules (PRMs) and therefore results in smaller and faster designs that potentially also reduce the power consumption. In this paper, we show that energy consumption is an important factor that has to be considered while implementing a parameterized design using DCS. In order to make a good choice for implementing a parameterized design with the goal of power optimized implementation, it is important to have a good power consumption estimate of the Dynamic Circuit Specialization. In this context, our paper presents a detailed investigation of the power consumption of a parameterized design implemented using DCS on the Xilinx Zynq-SoC FPGA. We propose an energy analysis of DCS and investigate the benefits of the use of DCS in comparison with a classic static FPGA implementation. We see that the power needed for the reconfiguration is much higher than the gain in power using the reconfiguration over the static implementation. An important reason is because of the CPU involved during the reconfiguration and the interface between the AXI bus and the HWICAP. To reduce the reconfiguration power, we include a clock gating technique to the reconfiguration interface AXIHWICAP that makes DCS more power efficient. We also relate the power gain to the size of the implementation and to the allowed time to reconfigure versus the useful run time. We conclude that for an implementation with 10 FIR filters, the reconfiguration time should not take more than 30.3% of the total time in order to remain energy efficient. Considering a specific use case with 10 FIR filters at a reconfiguration rate of 0.01, the energy consumption using DCS implementation is 20.5% lower than using the static FIR.
- 1 December 2015
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
MiCAP: a custom reconfiguration controller for dynamic circuit specialization
Field Programmable Gate Arrays (FPGAs) belong to a class of semiconductor devices whose hardware can be changed according to our needs. The configuration data (bitstreams) of an FPGA define the functionality of the FPGA. Therefore, a user can design the hardware and change it by modifying the bitstreams for a given set of requirements. One way of doing this is using Dynamic Circuit Specialization (DCS), an FPGA implementation technique that is optimized for a parameterized design. A design is said to be parameterized if some of its inputs are infrequently changing compared to the rest. In the DCS technique, for every change in parameterized input values, a new specialized circuit is generated during run-time and the FPGA is reconfigured accordingly. The time taken to reconfigure the FPGA with a specialized circuit is called reconfiguration time and is a major overhead of the DCS technique. To reduce this overhead, we propose an efficient custom reconfiguration controller built with a simple architecture which is customized to implement DCS. Our results indicate an increase in the reconfiguration speed by ≈ 17% and the FPGA resource utilization is reduced by ≈ 50% compared to the standard Xilinx reconfiguration controller.
- 1 December 2015
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
MiCAP: a custom reconfiguration controller for dynamic circuit specialization
Field Programmable Gate Arrays (FPGAs) belong to a class of semiconductor devices whose hardware can be changed according to our needs. The configuration data (bitstreams) of an FPGA define the functionality of the FPGA. Therefore, a user can design the hardware and change it by modifying the bitstreams for a given set of requirements. One way of doing this is using Dynamic Circuit Specialization (DCS), an FPGA implementation technique that is optimized for a parameterized design. A design is said to be parameterized if some of its inputs are infrequently changing compared to the rest. In the DCS technique, for every change in parameterized input values, a new specialized circuit is generated during run-time and the FPGA is reconfigured accordingly. The time taken to reconfigure the FPGA with a specialized circuit is called reconfiguration time and is a major overhead of the DCS technique. To reduce this overhead, we propose an efficient custom reconfiguration controller built with a simple architecture which is customized to implement DCS. Our results indicate an increase in the reconfiguration speed by ≈ 17% and the FPGA resource utilization is reduced by ≈ 50% compared to the standard Xilinx reconfiguration controller.