International Association
for Cryptologic Research

We study satisfiability modulo the theory of finite fields and give a decision procedure for this theory. We implement our procedure for prime fields inside the cvc5 SMT solver. Using this theory, we construct SMT queries that verify the correctness of various zero knowledge proof compilers on various input programs. Our experiments show that our implementation is vastly superior to previous approaches (which encode field arithmetic using integers or bit-vectors).

Because of the rapid growth of Internet of Things (IoT), embedded systems have become an interesting target for experienced attackers. ESP32~\cite{tech-ref-man} is a low-cost and low-power system on chip (SoC) series created by Espressif Systems. The firmware extraction of such embedded systems is a real threat to the manufacturer as it breaks its intellectual property and raises the risk of creating equivalent systems with less effort and resources. In 2019, LimitedResults~\cite{LimitedResultsPown} published power glitch attacks which resulted in dumping secure boot and flash encryption keys stored in the eFuses of ESP32. Therefore, Espressif patched this vulnerability and then advised its customers to use ESP32-V3, which is an updated SoC revision. This new version is hardened against fault injection attacks in hardware and software as announced by Espressif~\cite{ESPpatch}. In this paper, we present for the first time a deep hardware security evaluation for ESP32-V3. The main goal of this evaluation is to extract the firmware encryption key stored in the eFuses. This evaluation includes Fault Injection (FI) and Side-Channel (SC) attacks. First, we use Electromagnetic FI (EMFI) in order to show that ESP32-V3 doesn't resist EMFI. However, by experimental results, we show that this version contains a revised bootloader compared to ESP32-V1, which hardens dumping the eFuse keys by FI. Second, we perform a full SC analysis on the AES accelerator of ESP32-V3. We show that an attacker with a physical access to the device can extract all the keys of the hardware AES-256 after collecting 60K power measurements during the execution of the AES block. Third, we present another SC analysis for the firmware decryption mechanism, by targeting the decryption operation during the power up. Using this knowledge, we demonstrate that the full 256-bit AES firmware encryption key, which is stored in the eFuses, can be recovered by SC analysis using 300K power measurements. Finally, we apply practically the firmware encryption attack on Jade hardware wallet \cite{jade}.

Recent years have witnessed a push to bring multi-party computation (MPC) to practice and make it accessible to the end user/programmer. Despite novel ideas, on frontend language design (e.g., Wysteria, Viaduct), backend protocol design and implementation (e.g., ABY, MOTION), or both (e.g., SPDZ), classical compiler optimizations remain largely under-utilized (if not completely unused) in MPC programming. A likely reason is that such optimizations are often applied on a middle-end intermediate representation such as SSA.

We put forth a methodology for an MPC programming compilation toolchain, which by mimicking the compilation methodology of standard imperative languages enables middle-end optimizations on MPC, yielding significant improvements. To this direction we devise an MPC circuit compiler that allows MPC programming in what is essentially Python, and inherits the structure (and therefore optimization opportunities) of the classical compilation pipeline. Our key conceptual contribution is advancing an intermediate language, which we call MPC-IR, that can be viewed as the analogue, in an MPC program’s compilation, of (enriched) SSA form. MPC-IR is a particularly appealing intermediate language as it allows backend-independent optimizations, a close analogy to machine independent optimizations in classical compilers. Demonstrating the power of our approach, we focus on a specific backend-independent optimization, SIMD-vectorization: We devise a novel classical-compiler-inspired automatic SIMD-vectorization on MPC-IR, which we show leads to significant speedup in circuit generation time and running time, as well as significant reduction in communication size and number of gates over the corresponding iterative schedule.

We implement and benchmark our compiler from a Python-like program to an optimized circuit that can be fed into an MPC backend (for our benchmarks we make use of the MOTION backend for MPC). We view our exhaustive benchmarks as both a way to validate our optimization and end-to-end compiler, and as a contribution, by itself, to a more complete benchmarks suite for MPC programming—such benchmarks suites are common in classical compilers.

We initiate a formal study of individual cryptography Informally speaking, an algorithm Alg is individual if in every implementation of Alg there always exists an individual user that has full knowledge of the cryptographic secrets S used by Alg. In particular, it should be infeasible to design implementations of this algorithm that would hide the secret S by distributing it between a group of parties using an MPC protocol, or via outsourcing it to a trusted execution environment.

This paper describes a formalization of the specification and the algorithm of the public key encryption scheme CRYSTALS-KYBER as well as the verification of its $\delta$-correctness and indistinguishability under chosen plaintext attack (IND-CPA) security proof. The algorithms and proofs were formalized with only minimal assumptions in a modular way to verify the proofs for all possible parameter sets. During the formalization in this flexible setting, problems in the correctness proof were uncovered. Furthermore, the security of CRYSTALS-KYBER under IND-CPA was verified using a game-based approach. As the security property does not hold for the original version of CRYSTALS-KYBER, we only show the IND-CPA security for the latest versions. The security proof was verified under the hardness assumption of the module Learning-with-Errors Problem. The formalization was realized in the theorem prover Isabelle and is foundational.

As the number of blockchain projects grows, efficient cross-chain interoperability becomes more necessary. A common cross-chain protocol is the two-way peg, which is typically used to transfer assets between blockchains and their sidechains. The criticality of cross-chain protocols require that they are designed with strong security models, which can reduce usability in the form of long transfer times. In this paper, we present Flyover, a repayment protocol to speed up the transfer of bitcoins over federated pegs by allowing untrusted liquidity providers to advance funds for the users. Transfer times are reduced because liquidity providers do not have the same security requirements as the underlying cross-chain protocol. We illustrate the Flyover protocol on the cross-chain interoperability protocol that connects Bitcoin to the RSK sidechain and show how Flyover can reduce transfer times without reducing security. In addition to this, Flyover extends the cross-chain protocol by allowing liquidity providers to make smart contract calls on RSK on behalf of the user.

The ChaCha20-Poly1305 AEAD scheme is being increasingly widely deployed in practice. Practitioners need proven security bounds in order to set data limits and rekeying intervals for the scheme. But the formal security analysis of ChaCha20-Poly1305 currently lags behind that of AES-GCM. The only extant analysis (Procter, 2014) contains a flaw and is only for the single-user setting. We rectify this situation. We prove a multi-user security bound on the AEAD security of ChaCha20-Poly1305 and establish the tightness of each term in our bound through matching attacks. We show how our bound differs both qualitatively and quantitatively from the known bounds for AES-GCM, highlighting how subtle design choices lead to distinctive security properties. We translate our bound to the nonce-randomized setting employed in TLS 1.3 and elsewhere, and we additionally improve the corresponding security bounds for GCM. Finally, we provide a simple yet stronger variant of ChaCha20-Poly1305 that addresses the deficiencies highlighted by our analysis.

We propose a single-tiered hybrid Proof-of-Work consensus protocol to encourage decentralization in bitcoin. Our new mechanism comprises coupled puzzles of which properties differ from each other; the one is the extant outsourceable bitcoin puzzle while the other is non-outsourceable. Our new protocol enables miners to solve either puzzle as they want; therefore, blocks can be generated by either puzzle. Our hybrid consensus can be successfully implemented in bitcoin, because it is backward-compatible with existing bitcoin mining equipment(more precisely, existing bitcoin mining ASICs)

Oblivious RAM (ORAM), introduced by Goldreich and Ostrovsky (J. ACM '96), is a primitive that allows a client to perform RAM computations on an external database without revealing any information through the access pattern. For a database of size $N$, well-known lower bounds show that a multiplicative overhead of $\Omega(\log N)$ in the number of RAM queries is necessary assuming $O(1)$ client storage. A long sequence of works culminated in the asymptotically optimal construction of Asharov, Komargodski, Lin, and Shi (CRYPTO '21) with $O(\log N)$ worst-case overhead and $O(1)$ client storage. However, this optimal ORAM construction is known to be secure only in the honest-but-curious setting, where an adversary is allowed to observe the access patterns but not modify the contents of the database. In the malicious setting, where an adversary is additionally allowed to tamper with the database, this construction and many others in fact become insecure.

In this work, we construct the first maliciously secure ORAM protocol with worst-case $O(\log N)$ overhead and $O(1)$ client storage assuming one-way functions, which are also necessary. By the $\Omega(\log N)$ ORAM lower bound, our construction is asymptotically optimal. We can also interpret our construction as an online memory checker that matches the bandwidth of the best known online memory checkers while additionally hiding the access pattern. To achieve this, we intricately interleave the ORAM construction of Asharov et al. with online and offline memory checking techniques.

Flow is a high-throughput blockchain with a dedicated step for executing the transactions in a block and a subsequent verification step performed by Verification Nodes. To enforce integrity of the blockchain, the protocol requires a component that prevents Verification Nodes from approving execution results without checking. In our preceding work, we have sketched out an approach called Specialized Proof of Confidential Knowledge (SPoCK). Using SPoCK, nodes can provide evidence to a third party that they both executed the same transaction sequence without revealing the resulting execution trace. The previous Flow white paper presented a basic implementation of such scheme. In this note, we introduce a new SPoCK implementation that is more concise and more efficient than the previous proposal. We first provide a formal generic description of a SPoCK scheme as well as its security definition. Then we propose a new construction of SPoCK based on the BLS signature scheme. We support the new scheme with its proof of security under the appropriate computation assumptions.

Encryption alone is not enough for secure end-to-end encrypted messaging: a server must also honestly serve public keys to users. Key transparency has been presented as an efficient solution for detecting (and hence deterring) a server that attempts to dishonestly serve keys. Key transparency involves two major components: (1) a username to public key mapping, stored and cryptographically committed to by the server, and, (2) an out-of-band consistency protocol for serving short commitments to users. In the setting of real-world deployments and supporting production scale, new challenges must be considered for both of these components. We enumerate these challenges and provide solutions to address them. In particular, we design and implement a memory-optimized and privacy-preserving verifiable data structure for committing to the username to public key store.

To make this implementation viable for production, we also integrate support for persistent and distributed storage. We also propose a future-facing solution, termed ''compaction'', as a mechanism for mitigating practical issues that arise from dealing with infinitely growing server data structures. Finally, we implement a consensusless solution that achieves the minimum requirements for a service that consistently distributes commitments for a transparency application, providing a much more efficient protocol for distributing small and consistent commitments to users. This culminates in our production-grade implementation of a key transparency system (Parakeet) which we have open-sourced, along with a demonstration of feasibility through our benchmarks.

The private heavy-hitters problem is a data-collection task where many clients possess private bit strings, and data-collection servers aim to identify the most popular strings without learning anything about the clients' inputs. The recent work of Poplar constructed a protocol for private heavy hitters but their solution was susceptible to additive attacks by a malicious server, compromising both the correctness and the security of the protocol.

In this paper, we introduce PLASMA, a private analytics framework that addresses these challenges by using three data-collection servers and a novel primitive, called verifiable incremental distributed point function (VIDPF). PLASMA allows each client to non-interactively send a message to the servers as its input and then go offline. Our new VIDPF primitive employs lightweight techniques based on efficient hashing and allows the servers to non-interactively validate client inputs and preemptively reject malformed ones.

PLASMA drastically reduces the communication overhead incurred by the servers using our novel batched consistency checks. Specifically, our server-to-server communication depends only on the number of malicious clients, as opposed to the total number of clients, yielding a $182\times$ and $235\times$ improvement over Poplar and other state-of-the-art sorting-based protocols respectively. Compared to recent works, PLASMA enables both client input validation and succinct communication, while ensuring full security. At runtime, PLASMA computes the 1000 most popular strings among a set of 1 million client-held 32-bit strings in 67 seconds and 256-bit strings in less than 20 minutes respectively.

The increasing popularity of blockchain technology has affected the way we view many fields related to computer science, with E-commerce being no exception. The distributed nature and transparency of blockchain-based systems is one of its main perks, but it also raises some issues when it comes to privacy. Zero-knowledge proofs are very powerful building blocks when it comes to building privacy-preserving protocols, so, naturally, they have attracted a lot of attention in the last years. Following the recent collapse of the very popular crypto exchange FTX, we believe it is important to analyse how such events can be prevented in the future. This paper aims to highlight solutions that use zero-knowledge to prove solvency.

Blockchain is a newly emerging technology, however, it has proven effective in many applications because it provides multiple advantages, mainly as it represents a trust system in which data is encrypted in a way that cannot be tampered with or forged. Because it contains many details such as smart contracts, consensus, authentication, etc. the blockchain is a fertile ground for researchers where they can continually improve previous versions of these concepts. This paper introduces a new multi-signature scheme based on RSA. This scheme is designed to reduce the blockchain's size and prevent known attacks and is also applicable in many other settings that require multi-signatures. Our scheme is in the plain public key model, which means nodes do not need to prove knowledge or possession of their private key. In which, whatever the number of signers, the final signature size is equal to $O(k)$ where $k$ is a security parameter and no interaction between signers is needed. To verify that a number of parties have signed a shared message $m$, a verifier needs the signature, list of signers, and the message $m$. The presented practical short accountable-subgroup multi-signature (ASM) scheme allows a valid signature to disclose which subset generated the signature. It is worth noting that our multi-signatures with public key aggregation is an interactive two-round protocol and a multi-signature model applied to the entire block and not to individual transactions.

We give a construction of a 2-round blind signature scheme based on the hardness of standard lattice problems (Ring/Module-SIS/LWE and NTRU) with a signature size of 22 KB. The protocol is round-optimal and has a transcript size that can be as small as 60 KB. This blind signature is around $4$ times shorter than the most compact lattice-based scheme based on standard assumptions of del Pino and Katsumato (Crypto 2022) and around $2$ times shorter than the scheme of Agrawal et al. (CCS 2022) based on their newly-proposed one-more-SIS assumption. We also give a construction of a ``keyed-verification'' blind signature scheme in which the verifier and the signer need to share a secret key. The signature size in this case is only $48$ bytes, but more work needs to be done to explore the efficiency of the protocol which generates the signature.

Masking has become one of the most effective approaches for securing hardware designs against side-channel attacks. Irrespective of the effort put into correctly implementing masking schemes on a field programmable gate array (FPGA), leakage can be unexpectedly observed. This is due to the fact that the assumption underlying all masked designs, i.e., the leakages of different shares are independent of each other, may no longer hold in practice. In this regard, extreme temperatures have been shown to be an important factor in inducing leakage, even in correctly-masked designs. This has previously been verified using an external heat generator (i.e., a climate chamber). In this paper, we examine whether the leakage can be induced using the circuit components themselves. Specifically, we target masked neural networks (NNs) in FPGAs, with one of the main building blocks being block random access memory (BRAM) and flip-flops (FFs). In this respect, thanks to the inherent characteristics of NNs, our novel internal heat generators leverage solely the memories devoted to storing the user’s input, especially when frequently writing alternating patterns into BRAMs and FFs. The possibility of observing first-order leakage is evaluated by considering one of the most recent and successful first-order secure masked NNs, namely ModuloNET. ModuloNET is specifically designed for FPGAs, where BRAMs are used for storing the inputs and intermediate computations. Our experimental results demonstrate that undesirable first-order leakage can be observed by increasing the temperature when an alternating input is applied to the masked NN. To give a better understanding of the impact of extreme heat, we further perform a similar test on the design with FFs storing the input, where the same conclusion can be drawn.

The threat of chip-level tampering and its detection is a widely researched field. Hardware Trojan insertions are prominent examples of such tamper events. Altering the placement and routing of a design or removing a part of a circuit for side-channel leakage/fault sensitivity amplification are other instances of such attacks. While semi- and fully-invasive physical verification methods can confidently detect such stealthy tamper events, they are costly, time-consuming, and destructive. On the other hand, virtually all proposed non-invasive side-channel methods suffer from noise and, therefore, have low confidence. Moreover, they require activating the tampered part of the circuit (e.g., the Trojan trigger) to compare and detect the modification. In this work, we introduce a general non-invasive post-silicon tamper detection technique applicable to all sorts of tamper events at the chip level without requiring the activation of the malicious circuit. Our method relies on the fact that all classes of physical modifications (regardless of their physical, activation, or action characteristics) alter the impedance of the chip. Hence, characterizing the impedance can lead to the detection of the tamper events. To sense the changes in the impedance, we deploy known RF tools, namely, scattering parameters, in which we inject sine wave signals with high frequencies to the power distribution network (PDN) of the system and measure the “echo” of the signal. The reflected signals in various frequency bands reveal different tamper events based on their impact size on the die. To validate our claims, we performed extensive measurements on several proof-of-concept tampered hardware implementations realized on an FPGA manufactured with a 28 nm technology. Based on these groundbreaking results, we demonstrate that stealthy hardware Trojans, as well as sophisticated modifications of P&R, can be detected with high confidence.

We consider multi-party information-theoretic private computation. Such computation inherently requires the use of local randomness by the parties, and the question of minimizing the total number of random bits used for given private computations has received considerable attention in the literature.

In this work we are interested in another question: given a private computation, we ask how many of the players need to have access to a random source, and how many of them can be deterministic parties. We are further interested in the possible interplay between the number of random sources in the system and the total number of random bits necessary for the computation.

We give a number of results. We first show that, perhaps surprisingly, $t$ players (rather than $t+1$) with access to a random source are sufficient for the information-theoretic $t$-private computation of any deterministic functionality over $n$ players for any $t
We then turn to the question of the possible interplay between the number of random sources and the necessary number of random bits. Since for only very few settings in private computation meaningful bounds on the number of necessary random bits are known, we consider the AND function, for which some such bounds are known. We give a new protocol to $1$-privately compute the $n$-player AND function, which uses a single random source and $6$ random bits tossed by that source. This improves, upon the currently best known results (Kushilevitz et al., TCC'19), at the same time the number of sources and the number of random bits (KOPRT19 gives a $2$-source, $8$-bits protocol). This result gives maybe some evidence that for $1$-privacy, using the minimum necessary number of sources one can also achieve the necessary minimum number of random bits. We believe however that our protocol is of independent interest for the study of randomness in private computation.

This Paper proposes FssNN, a communication-efficient secure two-party computation framework for evaluating privacy-preserving neural network via function secret sharing (FSS) in semi-honest adversary setting. In FssNN, two parties with input data in secret sharing form perform secure linear computations using additive secret haring and non-linear computations using FSS, and obtain secret shares of model parameters without disclosing their input data. To decrease communication cost, we split the protocol into online and offline phases where input-independent correlated randomness is generated in offline phase while only lightweight ``non-cryptographic'' computations are executed in online phase. Specifically, we propose $\mathsf{BitXA}$ to reduce online communication in linear computation, DCF to reduce key size of the FSS scheme used in offline phase for nonlinear computation. To further support neural network training, we enlarge the input size of neural network to $2^{32}$ via ``MPC-friendly'' PRG.

We implement the framework in Python and evaluate the end-to-end system for private training between two parties on standard neural networks. FssNN achieves on MNIST dataset an accuracy of 98.0%, with communication cost of 27.52GB and runtime of 0.23h per epoch in the LAN settings. That shows our work advances the state-of-the-art secure computation protocol for neural networks.

We put forth a new cryptographic primitive for securely computing inner-products in a scalable, non-interactive fashion: any party can broadcast a public (computationally hiding) encoding of its input, and store a secret state. Given their secret state and the other party's public encoding, any pair of parties can non-interactively compute additive shares of the inner-product between the encoded vectors.

We give constructions of this primitive from a common template, which can be instantiated under either the LPN (with non-negligible correctness error) or the LWE (with negligible correctness error) assumptions. Our construction uses a novel twist on the standard non-interactive key exchange based on the Alekhnovich cryptosystem, which upgrades it to a non-interactive inner product protocol almost for free. In addition to being non-interactive, our constructions have linear communication (with constants smaller than all known alternatives) and small computation: using LPN or LWE with quasi-cyclic codes, we estimate that encoding a length-$2^{20}$ vector over a 32-bit field takes less that 2s on a standard laptop; decoding amounts to a single cheap inner-product.

We show how to remove the non-negligible error in our LPN instantiation using a one-time, logarithmic-communication preprocessing. Eventually, we show to to upgrade its security to the malicious model using new sublinear-communication zero-knowledge proofs for low-noise LPN samples, which might be of independent interest.