Dynamic Power Analysis of Standard-Cell FPGA Fabrics
Bo Bao, Jason Helge Anderson
Abstract
Bo Bao, Jason Helge Anderson
Abstract
Synthesizable FPGA fabrics offer several advantages over the full-custom FPGAs produced by the leading commercial FPGA vendors, including process portability and ease of customization to a particular application. In this work, we consider the dynamic power consumption of standard-cell synthesizable FPGAs and quantify the "gap" in power between a synthesized FPGA and its full-custom equivalent. An Intel Stratix-IV-like FPGA, targetable by the VTR flow [9], is implemented in 45nm standard cells using an ASIC toolflow. Post-layout RC extraction is performed, permitting an accurate delay-based simulation for a set of application benchmarks, and detailed power analysis using PrimeTime PX (PTPX). Power results are compared with the same benchmarks implemented on the commercial Stratix-IV (40nm technology). Results show that glitches are a more significant component of power in the standard-cell vs. full-custom FPGA, and that for sequential circuits, the dynamic power gap ranges from 1.3-3.3×.
A significance statement is not available in the OpenAlex record.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Synthesizable FPGA fabrics offer several advantages over the full-custom FPGAs produced by the leading commercial FPGA vendors, including process portability and ease of customization to a particular application. In this work, we consider the dynamic power consumption of standard-cell synthesizable FPGAs and quantify the "gap" in power between a synthesized FPGA and its full-custom equivalent. An Intel Stratix-IV-like FPGA, targetable by the VTR flow [9], is implemented in 45nm standard cells using an ASIC toolflow. Post-layout RC extraction is performed, permitting an accurate delay-based simulation for a set of application benchmarks, and detailed power analysis using PrimeTime PX (PTPX). Power results are compared with the same benchmarks implemented on the commercial Stratix-IV (40nm technology). Results show that glitches are a more significant component of power in the standard-cell vs. full-custom FPGA, and that for sequential circuits, the dynamic power gap ranges from 1.3-3.3×.
Key concepts: Stratix, Field-programmable gate array, Software portability, Computer science, Embedded system, Application-specific integrated circuit, Power analysis, Standard cell