M3-Mix: A Multi-Coin, Memory-Light, Mixer Architecture for Privacy-Preserving Embedded Devices

Abstract

Embedded systems, as computational platforms with limited memory, compute capability, and bandwidth-constrained interfaces, are deployed across industrial infrastructures such as smart grid metering units, and supply chain modules to manage transactional credits bound to physical commodities or operational entitlements. To facilitate decentralized coordination across administrative domains of embedded systems, blockchain-based ledgers provide globally verifiable settlement and token-based credits, yet their inherent transparency exposes transaction metadata: credit transaction flow, credit types and value transferred and exchanged, etc., thus raising confidentiality concerns. These concerns are compounded by the requirement to enable multi-type credit transfer and exchange across heterogeneous embedded systems, each backed by distinct blockchains, and by the challenge of coordinating cryptographic protocol execution where only edge servers, connected to the local embedded devices, maintain as blockchain nodes. We present M3-Mix, a system with protocols suite for cross-chain credit transfer and exchange that prevents source-destination traceability (mixer for unlinkability) and supports multi-token-type, mixing credit exchange under embedded system constraints via embedding devices’ cryptographic offloading to edge servers as blockchain relayers. The architecture integrates SNARK-verifiable nullifier-based mixing, commitment-compatible credit encoding, and a Book-Settle coordination protocol that resolves nondeterminism in token exchanges under adversarial reordering. We implement a comprehensive system using Groth16 over the hash within R1CS via the Gnark framework, and benchmark through emulated ARM embedded processors, demonstrating low-latency proof generation, small and bounded RAM usage, and constant-size proofs consisting of eight 256-bit words, using smart contract deployments on Ethereum Sepolia confirming practical gas efficiency with various number of transactions in a mixer obfuscated set (Merkle depths).