Nonmetal Platforms

Chips based on the Nonmetal group are designed for embedded tasks. They always come without MMU and are designed for Zephyr RTOS.

Platforms are ordered by atomic number, which reflects their capability tier. Each platform is a strict superset of the one before it.

Feature Overview

All platforms use a VexiiRiscv CPU. Higher-tier variants add caches, branch prediction, and wider execution pipelines for increased performance.

CPU

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

I-Cache

D-Cache

Branch prediction

Dual-issue

FPU

Privilege levels

M

M

M+U

M+U

M+U

M+U

RISC-V Extensions

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

M (multiply/divide)

A (atomics)

F (single-precision FPU)

D (double-precision FPU)

C (compressed)

Zba/Zbb/Zbs (bit manipulation)

Zicsr

Zifencei

Every platform includes on-chip SRAM and boots from external SPI Flash in XIP mode. HyperRAM is available starting with Nitrogen, and tightly-coupled memory with Oxygen.

Memory

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

On-chip SRAM

TCM

SPI Flash XIP

HyperRAM

Simpler variants use a shared bus, while higher-tier platforms switch to a crossbar for better throughput.

Interconnect

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

Shared bus

Crossbar

A DMA controller is available starting with Oxygen.

DMA

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

Channels

TBD

TBD

TBD

All platforms include a PLIC for interrupt handling.

Interrupt Controller

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

PLIC sources

JTAG debug is available on every platform. Higher-tier variants add trace support for real-time instruction tracing.

Debug

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

JTAG

Trace

Hydrogen and Carbon run on a single clock domain. Advanced power management for higher-tier platforms is planned.

Power

Hydrogen

Carbon

Nitrogen

Oxygen

Phosphorus

Sulfur

Clock domains

1

1

4

TBD

TBD

TBD

Performance

CoreMark 1.0, measured on FPGA with the port in software/<soc>/coremark. Scores are dominated by memory and ISA rather than by the pipeline: Hydrogen has no instruction cache and no hardware multiply, so every fetch is a SPI flash transaction and every multiply a libgcc call. Carbon adds both, which accounts for almost all of its advantage. Nitrogen’s further gain comes from the data cache, branch prediction and HyperRAM, and is much smaller because the core is no longer starved for instructions.

CoreMark

CoreMark/MHz

Iterations/s

Clock

ISA

Executed from

Hydrogen

0.005

0.24

50 MHz

rv32i_zicsr_zifencei

XIP flash

Carbon

1.590

79.48

50 MHz

rv32imc_zicsr_zifencei

XIP flash

Nitrogen

2.360

70.80

30 MHz

rv32imc_zicsr_zifencei_zicntr_zihpm

HyperRAM

Oxygen

Phosphorus

Sulfur

Built with GCC 14.3.0 and -O2 -march=<isa> -mabi=ilp32, MEM_METHOD=MEM_STACK, iterations auto-scaled so every run exceeds the ten seconds EEMBC requires. The ports have no floating point, so CoreMark’s own Iterations/Sec line is computed from a truncated whole-second time; the figures above are derived from the raw tick count and timer frequency that portable_fini() prints.

Nitrogen additionally reports mcycle and minstret, giving 423,714 cycles and 313,653 instructions per iteration, an IPC of 0.74. Its cycle count matches the machine timer to seven digits, confirming that the CPU and the timer share one clock.

See the available platforms for detailed specifications and usage instructions.