Mojo-V: Secret Computation for RISC-V

Mojo-V (pronounced “mojo-five”) is a new RISC-V extension that introduces privacy-oriented programming capabilities for RISC-V. Mojo-V implements secret computation, enabling secure, efficient, and data-oblivious execution without reliance on fragile software and programmer trust. By sequestering sensitive data in dedicated secret registers and encrypting memory under a third-party key, Mojo-V prevents disclosures and enforces computation that is both blind (no direct disclosures) and silent (no side channel leakage). The design integrates seamlessly into the existing RISC-V ISA with only a mode bit and four new instructions, enforced entirely at decode. Early results show near-native execution speeds while offering over 5-7 orders of magnitude performance improvement compared to fully homomorphic encryption (FHE), with a clear roadmap for integration into CPUs, GPUs, and specialized accelerators.

To learn more...

• Here is an intro video describing Mojo-V: https://www.youtube.com/watch?v=HUT46TcNyyM
• Slides that give an overview of the Mojo-V project: https://drive.google.com/file/d/1VVzZqYHvQgnKMgXZjg7I_cX2GzF7awSN

The current Mojo-V ISA Extension Specification (release 1.02):

• [In PDF format.] (https://drive.google.com/file/d/1yfiBqp0xyXD-S_G5d2o7Ggta5wI6nRot)

To contact the developers of Mojo-V:

• Email: [email protected]

🧩 Mojo-V Reference Platform — Release 1.03

🚧 Project Status

The Mojo-V Reference Platform release 1.03 implements a reference Spike simulator platform for RISC-V RV64GC and the EXO compiler for the Mojo-V ISA Specification v1.02. The current implementation supports fast, strong, and proof-carrying encryption modes, along with safe disclosure of encrypted computation results and certified random number generation. The release includes a wide range of Mojo-V tests, privacy-oriented benchmarks, and demonstrator applications for safe disclosure. It also includes the complete Mojo-V ISA specification and developer documentation.

This release is appropriate for use as i) a Mojo-V application development platform, ii) a golden model for validating Mojo-V hardware implementations, and iii) a reference implementation for security analysis. Current work focuses on the development of i) an LLVM-based Mojo-V compiler, ii) a gem5-based Mojo-V model for architectural exploration and analysis, and iii) a reference CVA6 SystemVerilog RTL implementation of RISC-V RV64GC with Mojo-V extensions.

Specification Version: 1.02 (August 2026)
Contact: [email protected]

Current components

1. Mojo-V ISA Spec v1.02

   • released in doc/

2. Spike (Instruction Set Simulator) with Mojo-V Extensions

   • Mojo-V integrated into riscv-isa-sim, and feature-complete for an RV64GC CPU with ML-KEM-512 key encapsulation for data contract loading, and SIMON-128 symmetric key encryption for secret computation protection.
   • To run Spike with Mojo-V extensions enabled, add the --isa=rv64gc_zicond_zkmojov_zicntr flag when running spike

3. Data Contract Multi-tool

   Data contracts are encrypted packets that allow a Mojo-V CPU's hardware to access the data access key and configuration information (e.g., memory encryption mode) for a Mojo-V encrypted data set. The DC Multi-tool enables the following capabilities:

   • Hardware developers can create public/private ML-KEM512 key pairs: public keys are shared with service providers, private keys are embedded into the Mojo-V hardware implementation.
   • Data owners can create data contracts and encrypt them under the public ML-KEM512 keys of service providers. The matching Mojo-V hardware can then perform secret computation on the protected 3rd-party encrypted data.

4. Mojo-V Bringup-Bench Benchmarks

   • Full battery of security tests for RV64GC+Mojo-V
   • Full battery of integrity attack tests for RV64GC+Mojo-V
   • Full battery of EXO compiler library tests
   • Full batteries of EXO math and string library tests
   • Numerous privacy-oriented benchmarks build using the EXO Mojo-V compiler library
   • Full battery of safe disclosure demonstration applications
   • Full battery of certified TRNG demonstration applications
   • Hand-coded examples (e.g., bubble-sort) showing Mojo-V working secret computation

Note, the remainder of the Bringup-bench benchmarks have NOT been ported to Mojo-V, as yet.

⚙️ Building and Running the Mojo-V Reference Platform

A. Install a RISC-V LLVM Compiler

You’ll need an LLVM-based RISC-V cross-compiler capable of producing RV64GC binaries.

Here is a good place to start: https://github.com/openssl/openssl

B. Install OpenSSL version 3.6 or newer

You’ll need a developer's installation of OpenSSL version 3.6 or newer. This provides libraries that implement ML-KEM512, used by Spike for protected key exchange.

Here is a good place to start: https://clang.llvm.org/get_started.html

C. Clone the Mojo-V Repository

git clone https://github.com/toddmaustin/mojo-v.git
cd mojo-v

D. Build the RISC-V Spike simulator with Mojo-V Support

sudo apt-get install device-tree-compiler libboost-regex-dev libboost-system-dev
cd riscv-isa-sim
mkdir build
cd build
../configure --prefix=$RISCV
make

E. Build and test the Data Contract Multi-tool

Data contracts are encrypted packets that allow a Mojo-V CPU's hardware to access the data access key and configuration information (e.g., memory encryption mode) for a Mojo-V encrypted data set.

cd dc-tool
make clean build test

E. Build and Run Mojo-V Bringup-Bench Benchmark Tests

1. Build the Spike device driver

   cd bringup-bench/target
   make
   

2. Configure your compiler

   Edit ../Makefile and set TARGET_CC for the mojov target to the location of your LVM Clang-based RISC-V compiler.

3. Build and test the Bringup-Bench test programs

   cd ..                # go to the top-level bringup-bench directory
   make TARGET=mojov-spike mojov-tests     # run all Mojo-V tests
   

   As an alternative, you can run an individual benchmark by going into its directory and running the following command.

   cd ../mojov-test
   make TARGET=mojov-spike clean build test
   

🧪 Mojo-V Bringup-Bench Tests Overview