International Association
for Cryptologic Research

Privacy and security are prime obstacles to the wider adoption of machine learning services offered by cloud computing providers. Namely, trusting users' sensitive data to a third-party infrastructure, vulnerable to both external and internal malicious attackers, restricts many companies from leveraging the scalability and flexibility offered by cloud services. We propose Soteria, a system for distributed privacy-preserving machine learning that combines the Apache Spark system, and its machine learning library (MLlib), with the confidentiality features provided by Trusted Execution Environments (e.g., Intel SGX). Soteria supports two main designs, each offering specific guarantees in terms of security and performance. The first encapsulates most of the computation done by Apache Spark on a secure enclave, thus offering stronger security. The second fine-tunes the Spark operations that must be done at the secure enclave to reduce the needed trusted computing base, and consequently the performance overhead, at the cost of an increased attack surface. An extensive evaluation of Soteria, with classification, regression, dimensionality reduction, and clustering algorithms, shows that our system outperforms state-of-the-art solutions, reducing their performance overhead by up to 41%. Moreover, we show that privacy-preserving machine learning is achievable while providing strong security guarantees.

Division properties, introduced by Todo at Eurocrypt 2015, are extremely useful in cryptanalysis, are an extension of square attack (also called saturation attack or integral cryptanalysis). Given their im- portance, a large number of works tried to offer automatic tools to find division properties, primarily based on MILP or SAT/SMT. This paper studies better modeling techniques for finding division properties using the Constraint Programming and SAT/SMT-based automatic tools. We use the fact that the Quine-McCluskey algorithm produces a concise CNF representation corresponding to the division trail table of an Sbox. As a result, we can offer significantly more compact models, which allow SAT and Constraint Programming tools to outperform previous results. To show the strength of our new approach, we look at the NIST lightweight candidate KNOT and Ascon. We show several new distinguishers with a lower data complexity for 17-round KNOT-256, KNOT-384 and 19- round KNOT-512. In addition, for the 5-round Ascon, we get a lower data distinguisher than the previous division-based results. Finally, we revisit the method to extend the integral distinguisher by composing linear layers at the input and output. We provide a formu- lation to find the optimal number of linear combinations that need to be considered. As a result of this new formulation, we prove that 18- round KNOT-256 and KNOT-384 have no integral distinguisher using conventional division property and we show this more efficiently than the previous methods.

We construct a constant-round composable protocol for blind and verifiable classical delegation of quantum computation, and show applications to secure quantum computation with classical communication. In particular, we give the first maliciously-secure multi-party protocols for BQP (bounded-error quantum polynomial-time) computation that only require a single party to have quantum capabilities. Assuming QLWE (the quantum hardness of learning with errors), we obtain the following. - A six-round protocol between one quantum server and multiple classical clients in the CRS (common random string) model. - A three-round protocol between one quantum server and multiple classical clients in the PKI (public-key infrastructure) + QRO (quantum random oracle) model. - A two-message protocol between quantum sender and classical receiver (a quantum non-interactive secure computation protocol), in the QRO model.

The only previously known approach for obtaining composable security of blind classical verification of quantum computation (Gheorghiu and Vidick, FOCS 2019) has inverse polynomial security and requires polynomially many rounds of interaction.

The property we require of classical verification of quantum computation that enables composability is malicious blindness, which stipulates that the prover does not learn anything about the verifier's delegated computation, even if it is able to observe whether or not the verifier accepted the interaction. To construct a protocol with malicious blindness, we use a classical verification protocol for sampBQP computation (Chung et al., Arxiv 2020), which in general has inverse polynomial soundness error, to prove honest evaluation of QFHE (quantum fully-homomorphic encryption) ciphertexts with negligible soundness error. Obtaining a constant-round protocol requires a strong parallel repetition theorem for classical verification of quantum computation, which we show following the "nearly orthogonal projector" proof strategy (Alagic et al., TCC 2020).

Anonymity networks, such as the Tor network, are highly decentralized and make heavy use of ephemeral identities. Both of these characteristics run in direct opposition to a traditional public key infrastructure, so entity authentication in an anonymity network can be a challenge. One system that Tor relies on is key-blinded signatures, which allow public keys to be transformed so that authentication is still possible, but the identity public key is masked. This is used in Tor during onion service descriptor lookup, in which a .onion address is resolved to a rendezvous point through which a client and an onion service can communicate. The mechanism currently used is based on elliptic curve signatures, so a post-quantum replacement will be needed.

We consider four fully post-quantum key-blinding schemes, and prove the unlinkability and unforgeability of all schemes in the random-oracle model. We provide a generic framework for proving unlinkability of key-blinded schemes by reducing to two properties, signing with oracle reprogramming and independent blinding. Of the four schemes, two are based on Round 3 candidates in NIST's post-quantum signature standardization process, Dilithium and Picnic. The other two are based on much newer schemes, CSI-FiSh and LegRoast, which have more favourable characteristics for blinding. CSI-FiSh is based on isogenies and boasts a very small public key plus signature sizes, and its group action structure allows for key-blinding in a straightforward way. LegRoast uses the Picnic framework, but with the Legendre symbol PRF as a symmetric primitive, the homomorphic properties of which can be exploited to blind public keys in a novel way. Our schemes require at most small changes to parameters, and are generally almost as fast as their unblinded counterparts, except for blinded Picnic, for which signing and verifying is roughly half as fast.

The best algorithms for the Learning Parity with Noise (LPN) problem require sub-exponential time and memory. This often makes memory, and not time, the limiting factor for practical attacks, which seem to be out of reach even for relatively small parameters. In this paper, we try to bring the state-of-the-art in solving LPN closer to the practical realm. We improve upon the existing algorithms by modifying the Coded-BKW algorithm to work under various memory constrains. We correct and expand previous analysis and experimentally verify our findings. As a result we were able to mount practical attacks on the largest parameters reported to date using only $2^{39}$ bits of memory.

We present a cryptographic Java library called Cryptimeleon designed for prototyping and benchmarking privacy-preserving cryptographic schemes. The library is geared towards researchers wanting to implement their schemes (1) as a sanity check for their constructions, and (2) for benchmark numbers in their papers. To ease the implementation process, Cryptimeleon "speaks the language" of paper writers. It offers a similar degree of abstraction as is commonly used in research papers. For example, bilinear groups can be used as the familiar black-box and Schnorr-style proofs can be described on the level of Camenisch-Stadler notation. It employs several optimizations (such as multi-exponentation) transparently, allowing the developer to phrase computations as written in the paper instead of having to conform to an artificial API for better performance.

Cryptimeleon implements (among others) finite fields, elliptic curve groups and pairings, hashing, Schnorr-style zero-knowledge proofs, accumulators, digital signatures, secret sharing, group signatures, attribute-based encryption, and other modern cryptographic constructions.

In this paper, we present the library, its capabilities, and explain important design decisions.

We introduce SPEEDY, a family of ultra low-latency block ciphers. We mix engineering expertise into each step of the cipher’s design process in order to create a secure encryption primitive with an extremely low latency in CMOS hardware. The centerpiece of our constructions is a high-speed 6-bit substitution box whose coordinate functions are realized as two-level NAND trees. In contrast to other low-latency block ciphers such as PRINCE, PRINCEv2, MANTIS and QARMA, we neither constrain ourselves by demanding decryption at low overhead, nor by requiring a super low area or energy. This freedom together with our gate- and transistor-level considerations allows us to create an ultra low-latency cipher which outperforms all known solutions in single-cycle encryption speed. Our main result, SPEEDY-6-192, is a6-round 192-bit block and 192-bit key cipher which can be executed faster in hardware than any other known encryption primitive (including Gimli in Even-Mansour scheme and the Orthros pseudorandom function) and offers 128-bit security. One round more, i.e., SPEEDY-7-192, provides full 192-bit security. SPEEDY primarily targets hardware security solutions embedded in high-end CPUs, where area and energy restrictions are secondary while high performance is the number one priority.

In the last decade, machine learning-based side-channel attacks became a standard option when investigating profiling side-channel attacks. At the same time, the previous state-of-the-art technique, template attack, started losing its importance and was more considered a baseline to compare against. As such, most of the results reported that machine learning (and especially deep learning) could significantly outperform the template attack. This does not mean the template attack does not have certain advantages even when compared to deep learning. The most significant one is that it does not have any hyperparameters to tune, making it easier to use.

We take another look at the template attack, and we devise a feature engineering phase allowing template attacks to compete or even outperform state-of-the-art deep learning-based side-channel attacks. More precisely, we show how a deep learning technique called the triplet model can be used to find highly efficient embeddings of input data, which can then be fed into the template attack resulting in powerful attacks.

Physical Unclonable Functions (PUFs) have been increasingly used as an alternative to non-volatile memory for the storage of cryptographic secrets. Research on side channel and fault attacks with the goal of extracting these secrets has begun to gain interest but no fault injection attack targeting the necessary error correction within a PUF device has been shown so far. This work demonstrates one such attack on a hardware fuzzy commitment scheme implementation and thus shows a new potential attack threat existing in current PUF key storage systems. After presenting evidence for the overall viability of the profiled attack by performing it on an FPGA implementation, countermeasures are analysed: we discuss the efficacy of hashing helper data with the PUF-derived key to prevent the attack as well as codeword masking, a countermeasure effective against a side channel attack. The analysis shows the limits of these approaches. In particular, it demonstrates the criticality of timing in codeword masking by confirming the attack's effectiveness on ostensibly protected hardware.

We give constructions of three-round secure multiparty computation (MPC) protocols for general functions that make black-box use of a two-round oblivious transfer. For the case of semi-honest adversaries, we make use of a two-round, semi-honest secure oblivious transfer in the plain model. This resolves the round-complexity of black-box (semi-honest) MPC protocols from minimal assumptions and answers an open question of Applebaum et al. (ITCS 2020). For the case of malicious adversaries, we make use of a two-round maliciously-secure oblivious transfer in the common random/reference string model that satisfies a (mild) variant of adaptive security for the receiver.

Single-trace attacks are a considerable threat to implementations of classic public-key schemes, and their implications on newer lattice-based schemes are still not well understood. Two recent works have presented successful single-trace attacks targeting the Number Theoretic Transform (NTT), which is at the heart of many lattice-based schemes. However, these attacks either require a quite powerful side-channel adversary or are restricted to specific scenarios such as the encryption of ephemeral secrets. It is still an open question if such attacks can be performed by simpler adversaries while targeting more common public-key scenarios.

In this paper, we answer this question positively. First, we present a method for crafting ring/module-LWE ciphertexts that result in sparse polynomials at the input of inverse NTT computations, independent of the used private key. We then demonstrate how this sparseness can be incorporated into a side-channel attack, thereby significantly improving noise resistance of the attack compared to previous works. The effectiveness of our attack is shown on the use-case of CCA2 secure Kyber $k$-module-LWE, where $k\in\{2,3,4\}$. Our $k$-trace attack on the long-term secret can handle noise up to a $\sigma \leq 1.2$ in the noisy Hamming weight leakage model, also for masked implementations. A $2k$-trace variant for Kyber1024 even allows noise $\sigma \leq 2.2$ also in the masked case, with more traces allowing us to recover keys up to $\sigma \leq 2.7$. Single-trace attack variants have a noise tolerance depending on the Kyber parameter set, ranging from $\sigma \leq 0.5$ to $\sigma \leq 0.7$. As a comparison, similar previous attacks in the masked setting were only successful with $\sigma \leq 0.5$.

We investigate the isogeny graphs of supersingular elliptic curves over \(\mathbb{F}_{p^2}\) equipped with a \(d\)-isogeny to their Galois conjugate. These curves are interesting because they are, in a sense, a generalization of curves defined over \(\mathbb{F}_p\), and there is an action of the ideal class group of \(\mathbb{Q}(\sqrt{-dp})\) on the isogeny graphs. We investigate constructive and destructive aspects of these graphs in isogeny-based cryptography, including generalizations of the CSIDH cryptosystem and the Delfs--Galbraith algorithm.

In this paper, we introduce Scabbard, a suite of post-quantum key-encapsulation mechanisms. Our suite contains three different schemes Florete, Espada, and Sable based on the hardness of module- or ring-learning with rounding problem. In this work, we first show how the latest advancements on lattice-based cryptography can be utilized to create new better schemes and even improve the state-of-the-art on post-quantum cryptography. We put particular focus on designing schemes that can optimally exploit the parallelism offered by certain hardware platforms and are also suitable for resource constrained devices. We show that this can be achieved without compromising the security of the schemes or penalizing their performance on other platforms. To substantiate our claims, we provide optimized implementations of our three new schemes on a wide range of platforms including general-purpose Intel processors using both portable C and vectorized instructions, embedded platforms such as Cortex-M4 microcontrollers, and hardware platforms such as FPGAs. We show that on each platform, our schemes can outperform the state-of-the-art in speed, memory footprint, or area requirements.

Searchable symmetric encryption (SSE) has attracted significant attention because it can prevent data leakage from external devices, e.g., clouds. SSE appears to be effective to construct such a secure system; however, it is not trivial to construct such a system from SSE in practice because other parts must be designed, e.g., user login management, defining the keyword space, and sharing secret keys among multiples users who usually do not have public key certificates. In this paper, we describe the implementation of two systems from the state-free dynamic SSE (DSSE) (Watanabe et al., ePrint 2021), i.e., a secure storage system (for a single user) and a chat system (for multiple users). In addition to the DSSE protocol, we employ a secure multipath key exchange (SXKEX) protocol (Costea et al., CCS 2018), which is secure against some classes of unsynchronized active attackers. It allows the chat system users without certificates to share a secret key of the DSSE protocol in a secure manner. To realize end-to-end encryption, the shared key must be kept secret; thus, we must consider how to preserve the secret on, for example, a user's local device. However, this requires additional security assumptions, e.g., tamper resistance, and it seems difficult to assume that all users have such devices. Thus, we propose a secure key agreement protocol by combing the SXKEX and login information (password) that does not require an additional tamper-resistant device. Combining the proposed key agreement protocol and the underlying state-free DSSE protocol allow users who know the password to use the systems on multiple devices.

Deep learning-based side-channel analysis already became a de-facto standard when investigating the most powerful profiling side-channel analysis. The results from the last few years show that deep learning techniques can efficiently break targets that are even protected with countermeasures. While there are constant improvements in making the deep learning-based attacks more powerful, little is done on evaluating such attacks' performance. Indeed, what is done today is not different from what was done more than a decade ago.

This paper considers how to evaluate deep learning-based side-channel analysis and whether the commonly used techniques give the best results. To that end, we consider different summary statistics and the influence of algorithmic randomness on the stability of profiling models. Our results show that besides commonly used metrics like guessing entropy, one should also show the standard deviation results to assess the attack performance properly. Our results show that using the arithmetic mean for guessing entropy does not yield the best results, and instead, a geometric mean should be used.

We revisit the popular adage that side-channel countermeasures must be combined to be efficient, and study its application to bitslice masking and shuffling. Our contributions are threefold. First, we improve this combination: by shuffling the shares of a masked implementation rather than its tuples, we can amplify the impact of the shuffling exponentially in the number of shares, while this impact was independent of the masking security order in previous works. Second, we evaluate the masking and shuffling combination's performance vs. security tradeoff under sufficient noise conditions: we show that the best approach is to mask first (i.e., fill the registers with as many shares as possible) and shuffle the independent operations that remain. Third, we discuss the challenges for implementing masking and shuffling under low noise conditions: we recall that such algorithmic countermeasures cannot be implemented securely without a minimum level of physical noise. We conclude that with moderate but sufficient noise, the bitslice masking + shuffling combination is relevant, and its interest increases when randomness is expensive and many independent operations are available for shuffling. When these conditions are not met, masking only is the best option. As a side result, we improve the best known attack against shuffling from Asiacrypt 2012, which we use in our concrete evaluations.

State-of-the-art re-keying schemes can be viewed as a tradeoff between efficient but heuristic solutions based on binary field multiplications, that are only secure if implemented with a sufficient amount of noise, and formal but more expensive solutions based on weak pseudorandom functions, that remain secure if the adversary accesses their output in full. Recent results on “crypto dark matter” (TCC 2018) suggest that low-complexity pseudorandom functions can be obtained by mixing linear functions over different small moduli. In this paper, we conjecture that by mixing some matrix multiplications in a prime field with a physical mapping similar to the leakage functions exploited in side-channel analysis, we can build efficient re-keying schemes based on “crypto-physical dark matter”, that remain secure against an adversary who can access noise-free measurements. We provide first analyzes of the security and implementation properties that such schemes provide. Precisely, we first show that they are more secure than the initial (heuristic) proposal by Medwed et al. (AFRICACRYPT 2010). For example, they can resist attacks put forward by Belaid et al. (ASIACRYPT 2014), satisfy some relevant cryptographic properties and can be connected to a “Learning with Physical Rounding” problem that shares some similarities with standard learning problems. We next show that they are significantly more efficient than the weak pseudorandom function proposed by Dziembowski et al. (CRYPTO 2016), by exhibiting hardware implementation results.

Due to the enormous energy consuming involved in the proof of work (POW) process, the resource-efficient blockchain system is urged to be released. The verifiable delay function (VDF), being slow to compute and easy to verify, is believed to be the kernel function of the next-generation blockchain system. In general, the reduction over a class group, involving many complex operations, such as the large-number division and multiplication operations, takes a large portion in the VDF. In this paper, for the first time, we propose a highspeed architecture for the reduction by incorporating algorithmic transformations and architectural optimizations. Firstly, based on the fastest reduction algorithm, we present a modified version to make it more hardware-friendly by introducing a novel transformation method that can efficiently remove the largenumber divisions. Secondly, highly parallelized and pipelined architectures are devised respectively for the large-number multiplication and addition operations to reduce the latency and the critical path. Thirdly, a compact state machine is developed to enable maximum overlapping in time for computations. The experiment results show that when computing 209715 reduction steps with the input width of 2048 bits, the proposed design only takes 137.652ms running on an Altera Stratix-10 FPGA at 100MHz frequency, while the original algorithm needs 3278ms when operating over an i7-6850K CPU at 3.6GHz frequency. Thus we have obtained a drastic speedup of nearly 24x over an advanced CPU.

Zilliqa Research Pte. Ltd. is the creator of Zilliqa – a scalable blockchain platform to run distributed applications. The company maintains the platform as well as develops tools and services around it.

The research team at Zilliqa consists of senior researchers and research engineers working on distributed systems, programming languages, system security and formal verification. Since starting in late 2017, the team has produced several scientific and engineering work ranging from distributed systems, e.g., the sharding design for scaling blockchains that is now being adopted by several other platforms; to programming languages, e.g., a new principled smart contract language amenable to formal verification called Scilla among others.

We firmly believe in conducting research with practical significance in the blockchain space and particularly research that can go into production systems like Zilliqa, Ethereum or Bitcoin.

Role Overview: As a blockchain researcher, you will be responsible for conducting impactful research that can improve the state of the art blockchain infrastructures and in particular the Zilliqa blockchain. In this role, you will be working very closely with the research team as well as system engineers. We offer a competitive salary, a creative work environment and an opportunity to push a research contribution into a production system.

Scope of Role: We are looking for a researcher with experience in blockchains, distributed systems and applied cryptography. The role will require the researcher to identify key research problems in the blockchain space and the Zilliqa platform in particular and conduct impactful research.

Experience:

Research experience either as a PhD or as research assistant.

Must have published at least 2 papers in top-tier conferences.

Must have a good engineering background.

Must have conducted research in distributed systems, applied cryptography, system security, and blockchains in general.

Applying for this Role: Contact us by sending your updated CV.

Closing date for applications:

Contact: amrit@zilliqa.com

The Cryptography and Privacy Engineering Group (ENCRYPTO) @CS Department @Technical University of Darmstadt offers a fully funded position as Doctoral Researcher (Research Assistant/PhD Student) in Private Machine Learning for Mobile Applications, ideally starting from Oct 1, 2021 for up to 3 years with the possibility of extension.
Job description: You'll work in the research training group/doctoral college Privacy & Trust for Mobile Users funded by the German Research Foundation (DFG). In our sub-project, we build cryptography-based private machine learning services for mobile applications and investigate their legal applicability (data protection) and economic feasibility in interdisciplinary collaborations. You conduct research, implement prototypes, and publish&present the results at top venues. You'll participate in teaching and supervise thesis students & student assistants.
We offer: We demonstrate that privacy is efficiently protectable in real-world applications via cryptographic protocols. Our open and international working environment facilitates excellent research in a sociable team. TU Darmstadt is a top research university for IT security, cryptography and CS in Europe. Darmstadt is a very international, livable and well-connected city in the Rhine-Main area around Frankfurt. Knowledge of German is beneficial, but not required, and TU Darmstadt offers corresponding support.
Your profile:

• Completed Master's degree (or equivalent) at a top university with excellent grades in IT security, computer science, or a similar area.
• Extensive knowledge in applied cryptography/IT security and very good software development skills. Knowledge in cryptographic protocols (ideally MPC) is a plus.
• Experience with/motivation for working with other disciplines, e.g., law or economics.
• Self-motivated, reliable, creative, can work independently, and want to do excellent research.
• Our working language is English: able to discuss/write/present scientific results in English. German is beneficial but not required.

Application deadline: Aug 21, 2021. Later applications are considered.

Closing date for applications:

Contact: Thomas Schneider (application@encrypto.cs.tu-darmstadt.de)

More information: https://encrypto.de/2021-RTG-EN