The Challenge
A European manufacturer of industrial communication gateways needed to evaluate post-quantum cryptography for their next product generation. The gateway connects field devices to cloud infrastructure and uses ECDH for key exchange. The design runs on a AMD Zynq UltraScale+ FPGA with limited remaining logic resources.
The engineering team had no prior PQC experience. Their questions:
- Does ML-KEM fit into the remaining FPGA resources alongside the existing design?
- What is the performance impact on connection setup latency?
- How is PQC integrated into the exisiting software and OS environment?
The Solution
A structured analysis, prototyping, and benchmarking effort demonstrated that ML-KEM can be efficiently integrated on the target FPGA, leading to a hybrid migration strategy with hardware-based PQC for key exchange, continued use of ECDH in the current generation, and a phased transition to full post-quantum cryptography aligned with IEC 62443 requirements.
Analysis
- Mapped all cryptographic usage across firmware and communication stack
- Identified ECDH key exchange as the primary quantum-vulnerable component
- Assessed available FPGA resources: ~12,000 LUTs and 30 BRAM blocks remaining
Prototyping
- Integrated KiviPQC-KEM (ML-KEM-768, Tiny variant) on the target Zynq UltraScale+ device
- Measured resource consumption:
- ~4000 LUTs
- 15 BRAM blocks
- Latency and speed measured and compared against requirements
- Validated coexistence with existing ECDH-based architecture
Migration Path
- Phase 1: Deploy ML-KEM for device-to-cloud key exchange (current product revision)
- Phase 2: Add ML-DSA for firmware signature verification (next product generation)
- Phase 3: Full PQC migration including certificate chain
- Provided documentation for customer's IEC 62443 security architecture review
The Results
The customer had a working ML-KEM prototype on their actual hardware after 4 weeks. Resource fit was confirmed, performance requirements were met, and the engineering team had a clear, prioritized migration roadmap. The decision to proceed with hardware-accelerated PQC was based on measured data, not estimates.
