Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QRAM: A Survey and Critique

Samuel Jaques, Arthur G. Rattew

TL;DR

This survey critically evaluates quantum random-access memory (QRAM) by distinguishing active (control-driven) and passive (ballistic) models and by surveying circuit-based and hardware-based proposals. It argues that passive, scalable QRAM faces fundamental obstacles, including routing/readout constraints, error propagation, and Hamiltonian-encoding challenges, making cheap large-scale passive QRAM unlikely. In contrast, circuit-based QRAM remains a practical though resource-intensive tool, with lower bounds and design trade-offs clarified for unary, bucket-brigade, and select-swap architectures. The analysis also shows that dequantization and strong parallel classical computation often erode hoped-for quantum advantages in quantum linear algebra and cryptanalytic tasks, though niche, noise-tolerant, or small-scale QRAM deployments may still offer practical value. Overall, the work provides a rigorous framework and toolkit for algorithm designers and hardware researchers to navigate QRAM's promises and limits while outlining open questions about error resilience and architectural feasibility.

Abstract

Quantum random-access memory (QRAM) is a mechanism to access data (quantum or classical) based on addresses which are themselves a quantum state. QRAM has a long and controversial history, and here we survey and expand arguments and constructions for and against. We use two primary categories of QRAM from the literature: (1) active, which requires external intervention and control for each QRAM query (e.g. the error-corrected circuit model), and (2) passive, which requires no external input or energy once the query is initiated. In the active model, there is a powerful opportunity cost argument: in many applications, one could repurpose the control hardware for the qubits in the QRAM (or the qubits themselves) to run an extremely parallel classical algorithm to achieve the same results just as fast. We apply these arguments in detail to quantum linear algebra and prove that most asymptotic quantum advantage disappears with active QRAM systems, with some nuance related to the architectural assumptions. Escaping the constraints of active QRAM requires ballistic computation with passive memory, which creates an array of dubious physical assumptions, which we examine in detail. Considering these details, in everything we could find, all non-circuit QRAM proposals fall short in one aspect or another. In summary, we conclude that cheap, asymptotically scalable passive QRAM is unlikely with existing proposals, due to fundamental obstacles that we highlight. These obstacles are deeply rooted in the requirements of QRAM, but are not provably inevitable; we hope that our results will help guide research into QRAM technologies that circumvent or mitigate these obstacles. Finally, circuit-based QRAM still helps in many applications, and so we additionally provide a survey of state-of-the-art techniques as a resource for algorithm designers using QRAM.

QRAM: A Survey and Critique

TL;DR

This survey critically evaluates quantum random-access memory (QRAM) by distinguishing active (control-driven) and passive (ballistic) models and by surveying circuit-based and hardware-based proposals. It argues that passive, scalable QRAM faces fundamental obstacles, including routing/readout constraints, error propagation, and Hamiltonian-encoding challenges, making cheap large-scale passive QRAM unlikely. In contrast, circuit-based QRAM remains a practical though resource-intensive tool, with lower bounds and design trade-offs clarified for unary, bucket-brigade, and select-swap architectures. The analysis also shows that dequantization and strong parallel classical computation often erode hoped-for quantum advantages in quantum linear algebra and cryptanalytic tasks, though niche, noise-tolerant, or small-scale QRAM deployments may still offer practical value. Overall, the work provides a rigorous framework and toolkit for algorithm designers and hardware researchers to navigate QRAM's promises and limits while outlining open questions about error resilience and architectural feasibility.

Abstract

Quantum random-access memory (QRAM) is a mechanism to access data (quantum or classical) based on addresses which are themselves a quantum state. QRAM has a long and controversial history, and here we survey and expand arguments and constructions for and against. We use two primary categories of QRAM from the literature: (1) active, which requires external intervention and control for each QRAM query (e.g. the error-corrected circuit model), and (2) passive, which requires no external input or energy once the query is initiated. In the active model, there is a powerful opportunity cost argument: in many applications, one could repurpose the control hardware for the qubits in the QRAM (or the qubits themselves) to run an extremely parallel classical algorithm to achieve the same results just as fast. We apply these arguments in detail to quantum linear algebra and prove that most asymptotic quantum advantage disappears with active QRAM systems, with some nuance related to the architectural assumptions. Escaping the constraints of active QRAM requires ballistic computation with passive memory, which creates an array of dubious physical assumptions, which we examine in detail. Considering these details, in everything we could find, all non-circuit QRAM proposals fall short in one aspect or another. In summary, we conclude that cheap, asymptotically scalable passive QRAM is unlikely with existing proposals, due to fundamental obstacles that we highlight. These obstacles are deeply rooted in the requirements of QRAM, but are not provably inevitable; we hope that our results will help guide research into QRAM technologies that circumvent or mitigate these obstacles. Finally, circuit-based QRAM still helps in many applications, and so we additionally provide a survey of state-of-the-art techniques as a resource for algorithm designers using QRAM.
Paper Structure (69 sections, 15 theorems, 32 equations, 11 figures, 1 table, 1 algorithm)

This paper contains 69 sections, 15 theorems, 32 equations, 11 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Summary
  3. Outline
  4. Definitions and Notation
  5. QRAM Definitions
  6. CRACM (Classical access classical memory):
  7. CRAQM (Classical access quantum memory):
  8. QRACM (Quantum access classical memory):
  9. QRAQM (Quantum access quantum memory)
  10. Routing and Readout
  11. Alternate Notation
  12. Related Work.
  13. Applications
  14. Optimizing calculations.
  15. Dihedral Hidden Subgroup.
  16. ...and 54 more sections

Key Result

Theorem 4.1

Given a degree $k$ polynomial on the interval $x \in [-1, 1]$, with $f(x) = \sum_{j=0}^k a_j x^j$, a $d$-sparse Hermitian matrix $H \in \mathbb{C}^{N\times N}$ and an $\ell_2$-normalized vector $\left\vert\psi\right\rangle\in\mathbb{C}^N$, no general quantum algorithm can prepare the state $f(H)\lef

Figures (11)

  • Figure 1: An illustration of a quantum computer in the memory peripheral model CRYPTO:JaqSch19. Classical controllers are the chips on the top, which send signals (in red) to qubits (in blue).
  • Figure 2: Circuit diagram for naive QRACM. Here $? \atop =j$ means a circuit that checks if the input equals $j$, and flips the target output if so.
  • Figure 3: Recursive controlled unary QRACM from PRX:BGB+2018.
  • Figure 4: Schematic of the overall tree structure of circuit bucket-brigade QRACM. In the tree structure in (a), the lines between nodes are only to help exposition and do not represent any physical device (unlike in hardware bucket-brigade QRACM; see Section \ref{['sec:bucket-brigade']}). (b) shows how each node consists of 2 qubits, and labels them to show where the circuits in (c) and (d) apply their gates.
  • Figure 5: Select-swap QRACM access. $T'$ is the table formed from $T$ by combining each page of $2^\ell$ consecutive bits of $T$ into one word in $T'$.
  • ...and 6 more figures

Theorems & Definitions (30)

  • Definition 1
  • Definition 2
  • Theorem 4.1: Quantum Polynomial Eigenvalue Transform
  • Theorem 4.2: Classical Parallel Polynomial Eigenvalue Transform
  • proof
  • Lemma 5.1
  • Theorem 5.2
  • proof
  • Definition 3
  • Lemma 6.1
  • ...and 20 more