TL;DR
This paper investigates the feasibility and impact of using quantum-resistant cryptography for securing software updates on low-power IoT devices. By benchmarking post-quantum signature schemes against current ones on various microcontrollers, the authors demonstrate that while quantum-resistant updates are achievable, they come with significant trade-offs in terms of resource usage and network transfer costs.
What Problem Does This Paper Address?
The Internet of Things (IoT) is rapidly expanding, with devices often having lifespans exceeding a decade. This longevity means that even if quantum computers are still a distant threat, current IoT devices could be vulnerable to attacks from future quantum adversaries. Software updates are a critical mechanism for maintaining IoT security, but the update process itself can be an attack vector. The IETF's SUIT standard aims to secure these updates using digital signatures and hash functions. However, existing SUIT performance evaluations have not considered the post-quantum era. This paper addresses this gap by analyzing the performance and resource implications of integrating quantum-resistant digital signatures into the SUIT framework for resource-constrained, low-power IoT devices.
Methods
The authors leverage the open-source RIOT operating system as a platform to study the SUIT workflow. They focus on two key cryptographic components: hash functions and digital signatures.
For hash functions, they analyze SHA-2 and SHA-3 (specifically SHA3-256) and their suitability for low-power IoT devices, considering factors like flash memory footprint and execution speed.
For digital signatures, they benchmark several post-quantum candidates against current pre-quantum schemes. The post-quantum schemes evaluated are:
- LMS (Leighton-Micali Hash-based Signatures): A hash-based signature scheme.
- Falcon: A lattice-based signature scheme.
- Dilithium: Another lattice-based signature scheme.
These are compared with pre-quantum schemes:
- Ed25519: An elliptic-curve signature scheme.
- ECDSA (with secp256 curve): Another widely used elliptic-curve signature scheme.
The benchmarks were conducted on a variety of IoT hardware, including ARM Cortex-M, RISC-V, and Espressif (ESP32) microcontrollers, to assess performance in terms of flash and RAM usage, signing and verification times, and stack memory requirements.
Datasets
Not discussed in the paper. The paper uses existing hardware platforms and software implementations for its benchmarks.
Key Insights and Ideas
- Quantum threat is real for long-lived IoT: Devices deployed today will likely outlive the security guarantees of current cryptography against future quantum computers.
- SUIT is adaptable, but with costs: The SUIT standard's modularity allows for the integration of post-quantum cryptography, but this comes with significant overheads.
- Post-quantum signatures are resource-intensive: Compared to current elliptic-curve schemes, post-quantum signatures (LMS, Falcon, Dilithium) have substantially larger public keys and signatures, demanding more flash memory and network bandwidth.
- Verification is faster, signing is slower: A surprising finding is that post-quantum signature verification can be significantly faster than pre-quantum schemes on some platforms, while signing is generally slower. This is a crucial trade-off for IoT devices where verification is the primary on-device operation.
- Hash function choice matters for efficiency: While SHA-2 and SHA-3 offer similar security, their implementations have different performance characteristics on microcontrollers, with SHA3-256 offering speed at the cost of larger flash footprints.
- No single "best" post-quantum signature: The optimal choice depends heavily on the specific IoT use case, particularly the size of the software update and the frequency of updates. Falcon shows promise for smaller updates due to its relatively smaller signature size compared to LMS and Dilithium.
Key Results
- Signature Size: Post-quantum signatures are 10-74x larger than Ed25519/ECDSA signatures. For example, LMS signatures are 4756 bytes compared to 64 bytes for Ed25519.
- Public Key Size: Post-quantum public keys are also significantly larger, ranging from 60 bytes for LMS to 1312 bytes for Dilithium, compared to 32 bytes for Ed25519.
- Verification Speed: On average, post-quantum signature verification was found to be faster than pre-quantum schemes on the tested hardware. For instance, Falcon verification was up to 2.65x faster than Ed25519 on the ARM Cortex-M.
- Signing Speed: Signing operations for post-quantum schemes were generally slower. The fastest post-quantum signing (Falcon) was up to 7.94x slower than Ed25519 (Monocypher).
- Flash Footprint: Post-quantum schemes can increase the flash memory footprint by over 11x compared to pre-quantum schemes.
- Network Transfer Costs: For software updates, the increased signature size leads to higher network transfer costs, ranging from a 1.1% increase for Falcon to a 4.3% increase for Dilithium (when including crypto in the transfer). For larger updates, this cost becomes less significant relative to the overall update size.
Limitations and Future Work
- Limited Hardware Scope: While the paper tested three representative microcontroller architectures (ARM Cortex-M, RISC-V, ESP32), the results might not generalize to all low-power IoT devices.
- Specific Implementations: The performance is dependent on the specific implementations of the cryptographic algorithms used. Further optimization could alter the results.
- Maturity of Schemes: The paper notes that some post-quantum schemes (e.g., isogeny-based, code-based) are still in early stages of development and security analysis, making them less suitable for immediate deployment.
- Future Work: The authors suggest prioritizing the stabilization of current post-quantum signature implementations and their integration into embedded software platforms like RIOT. They also highlight the ongoing NIST standardization process and the need for re-evaluation as new candidates emerge. Further research into optimizing these schemes for even more constrained environments is also recommended.
Summary updated