Small Hazard-free Transducers

TL;DR

This work addresses the challenge of hazard-free circuit design for transcription functions derived from finite-state transducers. By introducing a universal encoding that tracks possible state-set reachability under metastable inputs, the authors extend the parallel-prefix computation framework to hazard-free settings and prove that constant-size transducers admit hazard-free implementations with size $\mathcal{O}((\kappa^3 + (2^{\ell}/\ell)\kappa^2 + 2^{\ell}\kappa\lambda)n)$ and depth $\mathcal{O}(\log\kappa\log n + \ell)$, where $\kappa$ and $\lambda$ depend on the transducer size $|S|$, input encoding $\ell$, and output width $m$. They establish associativity for hazard-free matrix multiplication and show that the universal encoding preserves necessary information across function compositions, enabling efficient hazard-free prefix computations. The approach yields fixed-parameter tractability in $\max\{\ell,|S|\}$ and remains asymptotically optimal in the input length $n$, despite exponential overhead in the transducer size. This framework opens avenues for hazard-free synthesis of arithmetic circuits, including potential hazard-free addition, by providing a general, automated method to convert transducers into robust digital circuits under metastability.

Abstract

Ikenmeyer et al. (JACM'19) proved an unconditional exponential separation between the hazard-free complexity and (standard) circuit complexity of explicit functions. This raises the question: which classes of functions permit efficient hazard-free circuits? In this work, we prove that circuit implementations of transducers with small state space are such a class. A transducer is a finite state machine that transcribes, symbol by symbol, an input string of length $n$ into an output string of length $n$. We present a construction that transforms any function arising from a transducer into an efficient circuit of size $\mathcal{O}(n)$ computing the hazard-free extension of the function. More precisely, given a transducer with $s$ states, receiving $n$ input symbols encoded by $l$ bits, and computing $n$ output symbols encoded by $m$ bits, the transducer has a hazard-free circuit of size $2^{\mathcal{O}(s+\ell)} m n$ and depth $\mathcal{O}(s\log n + \ell)$; in particular, if $s, \ell,m\in \mathcal{O}(1)$, size and depth are asymptotically optimal. In light of the strong hardness results by Ikenmeyer et al. (JACM'19), we consider this a surprising result.

Figures (4)

• Figure 1: The shift transducer delays the input by one symbol. It serves as a running example.
• Figure 2: Steps $1$ to $3$ of the circuit implementing the transcription function. Enc denotes the computation of the universal encoding by the generic construction of jukna2021notes. Hazard-free Boolean matrix multiplication is denoted by $\mathcal{A}\cdot_\mathfrak{u}\mathcal{B}$.
• Figure 3: Step 4 of the circuit implementing the transcription function, assuming $m=1$, for the case that every preimage is encoded in the state vector (1) and the case that there is at least one preimage not encoded in the state vector (2). To increase readability we do not enumerate all elements of $\Sigma$ but use $\sigma$, $\sigma'$, and dots to denote that the step is repeated for every element of $\Sigma$.
• Figure 4: Extension of the first example transducer, that shifts to outputting ones only after reception of two consecutive ones. Vice versa the transducer shifts to outputting zeros upon reception of two consecutive zeros.

Theorems & Definitions (26)

• Definition 1.1: Superposition
• Definition 1.2: Resolution
• Definition 1.4: Hazard-free Extensions
• Definition 1.5: k-Bit Hazards
• Definition 1.6: Mealy Machine
• Definition 1.7: Transcription Function $\tau$
• Theorem 1.7
• Corollary 1.8
• Corollary 1.9
• Definition 3.1: Universal Function Encoding