The ESP website contains the most up-to-date information on the ESP project. The Documentation page contains detailed guides and video tutorials that will be released periodically to help users get the most out of ESP.
ESP is an open-source platform for heterogeneous SoC design and prototype on FPGA. It provides a flexible tile-based architecture built on a multi-plane network-on-chip.
In addition to the architecture, ESP provides users with templates and scripts to create new accelerators from SystemC, Chisel, and C/C++. The ESP design methodology eases the process of integrating processors and accelerators into an SoC by offering platform services, such as DMA, distributed interrupt, and run-time coherence selection, that hide the complexity of hardware and software integration from the accelerator designer.
Currently, ESP supports the integration of the LEON3 processor from GRLIB, the Ariane core from the PULP Platform, and the Ibex core from lowRISC. LEON3 implements the SPARC V8 32-bit ISA, Ariane implements the RISC-V 64-bit ISA, and Ibex implements the RISC-V 32-bit ISA.
In addition to processor cores, ESP embeds accelerator design examples created with Stratus HLS in SystemC, Vivado HLS in C/C++ and Chisel.
Furthermore, ESP can serve as a platform to integrate third-party IP blocks. For example, ESP integrates the NVIDIA Deep Learning Accelerator NVDLA, which can be placed on any ESP accelerator tile.
Overview paper:
Paolo Mantovani, Davide Giri, Giuseppe Di Guglielmo, Luca Piccolboni, Joseph Zuckerman, Emilio G. Cota, Michele Petracca, Christian Pilato, Luca P. Carloni. "Agile SoC Development with Open ESP." IEEE/ACM International Conference On Computer Aided Design (ICCAD), 2020.
The Publications page of the ESP website contains the complete list of publications related to ESP.
Here is a brief description of the main directories in the repository, please refer to the READMEs inside each of them for more information.
-
acceleratorscontains multiple accelerator design and integration flows, as well as many example accelerators. -
constraintscontains the constraints and attributes for each supported FPGA board (or ASIC technology). -
socscontains the working folders for launching all Make targets. There is one working folder for each supported FPGA board (or ASIC technology). -
rtlcontains the whole RTL code base, excluding the accelerators RTL and the RTL generated in the working folder by the SoCGen and SocketGen tools. -
softcontains bootloader, Linux kernel and root file system, and bare-metal library for each of the available processor cores. It also contains bare-metal, user space and kernel space libraries for invoking and managing accelerators. -
techis the destination of the RTL generated by the HLS-based and Chisel-based accelerator design flows. It is also the destination of the RTL generated with HLS for the SystemC implementation of the cache hierarchy. The generated RTL is organized based on the target FPGA (or ASIC) technology. -
toolscontains tools for design automation and for communicating with an ESP SoC from a host machine. -
utilscontains various scripts and utilities, including the main Makefiles, the RTL file lists, and the software toolchains installation scripts. -
.cachecaches some compiled libraries so they only need to be compiled once (e.g. Xilinx simulation libraries).
- Dynamic partial reconfiguration SoC flow
- Expanded support for ASIC design
- New platform services for programmable accelerators
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent