Full Documentation

A RISC-V 32-bit microcontroller written in VHDL targeted for an FPGA.

This RISC-V microcontroller uses the RV32IM instruction set and the Zicsr, Zicntr, Zicond, Zihpm, Zba, Zbs, Sdext and Sdtrig extensions. The microcontroller is build around a three-stage pipelined core.

All instructions from the "The RISC-V Instruction Set Manual Volume I, Unprivileged Architecture, Version 20240411", RV32, are supported. The instructions from the M standard (hardware multiply/divide), RV32, are supported. From the "The RISC-V Instruction Set Manual: Volume II Privileged Architecture, Version 20240411", RV32, ECALL, EBREAK, MRET and WFI are supported. The microcontroller supports Machine mode only. Traps (interrupts and exceptions) are supported.

Loads from memory require 3 clocks, stores require 2 clocks. CSR operations require 1 clock. Multiplies require 3 clocks, divides require 18 clocks. Jumps/branches taken require 3 clocks, the microcontroller does not support branch prediction. All other instructions require 1 clock.

Current Coremark testbench shows a CPI of 1.73 and a score of 1.93 coremark/MHz.

Software is written in C, (C++ is supported but there are some limitations) and compiled using the RISC-V GNU C/C++ compiler.

The design is equipped with bootloader and on-chip debugger hardware (both can be switched off). The bootloader can be used without the on-chip debugger to upload executables to the processor. The on-chip debugger can be used with OpenOCD and GDB. It works with Eclipse CDT.

The microcontroller uses FPGA onboard RAM blocks to emulate RAM and program ROM. There is no support for cache or external RAM. Programs are compiled with the GNU C compiler for RISC-V and the resulting executable is transformed to a VHDL synthesizable ROM table.

• ROM: a ROM of 64 kB is available (placed in onboard RAM, may be extended).

• BOOT: a bootloader ROM of 4 kB (placed in onboard RAM).

• RAM: a RAM of 32 kB using onboard RAM block available (may be extended).

• I/O: a simple 32-bit input and 32-bit output is available, as is a simple 7/8/9-bit UART with interrupt capabilities. Two SPI devices are available, with one device used for SD card socket (DE0-CV board, no interrupt) and a general purpose SPI device with hardware NSS. Two I2C devices are available. A simple timer with interrupt is provided. A more elaborate timer is included and can generate waveforms (Output Compare and PWM) or count edges (Input Capture). A watchdog timer is available, it can generate a system wide reset or an NMI. A machine mode software interrupt is provided. The external system timer MTIME is located in the I/O so it’s memory mapped.

ROM starts at 0x00000000, BOOT (if available) starts at 0x10000000, RAM starts at 0x20000000, I/O starts at 0xF0000000. May be changed on 256 MB (top 4 bits) sections.

A number of CSR registers are implemented: time, timeh, [m]cycle, [m]cycleh, [m]instret, [m]instreth, mvendorid, marchid, mimpid, mhartid, mstatus, mstatush, misa, mie, mtvec, mscratch, mepc, mcause, mip, mcountinhibit as are the HPM counters and event selectors. If on-chip debugging is enabled, the dcsr, dpc, tselect, tdata1, tdata2 and tinfo CSRs are available. Some of these CSRs are hardwired. Others will be implemented when needed. The time and timeh CSRs produces the time since reset in microseconds, shadowed from the External Timer memory mapped registers. Also two custom CSRs are implemented: mxhw which holds information of included peripherals and mxspeed which contains the synthesized clock speed.

The microcontroller is developed on a Cyclone V FPGA (5CEBA4F23C7) with the use of the DE0-CV board by Terasic and Intel Quartus Prime Lite 23.1. Simulation is possible with QuestaSim Intel Starter Edition. You need a (free) license for that. The processor uses about 2900 ALM (cells) of 18480, depending on the settings. In the default settings, ROM, BOOT, RAM and registers uses 43% of the available RAM blocks.

The design is also tested on a Digilent Arty S7/50 and a Cmod S7/25 (Spartan 7) board.

A number of C programs have been tested, created by the GNU C/C++ Compiler for RISC-V. We tested the use of (software) floating point operations (both float and double) and tested the mathematical library (sin, cos, et al.). Traps (interrupts and exceptions) are tested and work. Assembler programs can be compiled by the GNU assembler. We provide a CRT (C startup) and linker file. C++ is supported but many language concepts (e.g. cout with iostream) create a binary that is too big to fit in the ROM.

We provide a basic set of systems call, trapped (ECALL) and non-trapped (functions overriding the C library functions). Trapped system calls are by default set up by the RISC-V C/C++ compiler, so no extra handling is needed.

The processor is equipped with an on-chip debugger that complies to the RISC-V Debug Specification v1.0.0-rc3. It is an all-hardware solution, there is no program buffer involved. All processor registers (including CSRs) can be read and written (when possible). Memory can be read and written. There is one hardware breakpoint available. De on-chip debugger is compatible with OpenOCD and GDB. It can be used with Eclipse CDT. By default, the on-chip debugger is disabled.

By default, the design is equipped with a bootloader. When resetting the FPGA, the bootloader waits about 5 seconds @50 MHz before the program in the ROM is started. Using the bootloader, a program can written to the ROM (see the documentation). The bootloader can also be used to inspect the memory contents.

There is support for Windows tools. srec2vhdl and upload can be build with GCC MinGW and Visual Studio. For building the RISC-V programs, a RISC-V GNU GCC compiler is needed.

Best is to use a precompiled compiler for Windows and build tools (make, rm, mkdir etc.). Please have a look at xPack RISC-V Toolchain and Windows build tools. For building srec2vhdl and upload, you need a GCC native compiler. Have a look at The xPack GNU Compiler Collection (GCC). For on-chip debugging, see xPack OpenOCD.

• We are not planning the C standard.

• Implement clock stretching and arbitration in the I2C1/I2C2 peripherals.

• Adding input synchronization for SPI1/SPI2 peripherals.

• Implement an I/O input/output multiplexer for GPIOA PIN and POUT. This will enable I/O functions to be multiplexed with normal port I/O.

• Test more functions of the standard and mathematical libraries.

• It is not possible to print long long (i.e. 64-bit) using printf et al. When using the format specifier %lld, printf just prints ld. This due to lack of support in the nano library.

• Adding Zbb extension.

• To start the pre-programmed bootloader, make sure the UART1 RxD pin is connected to a serial device OR make sure this pin is pulled high (DE0-CV board).

This microcontroller is for educational purposes only. Work in progress. Things might change. Use with care.