Stab-QRAM: Clifford QRAM for Affine Boolean Data

• Stab-QRAM is a domain-specific quantum RAM using affine Boolean functions and Clifford gates for exact data loading.
• It achieves optimal CNOT scheduling via a bipartite interaction graph, yielding logarithmic logical depth and space efficiency.
• The design avoids non-Clifford overhead, providing a specialized oracle for optimization, quantum linear systems, and discrete dynamical systems.

Stabilizer-QRAM, usually abbreviated Stab-QRAM, is a domain-specific quantum random access memory architecture for classical data with affine Boolean structure,

$$f(\mathbf{x})=A\mathbf{x}+\mathbf{b},\qquad A\in\mathbb{F}_2^{m\times n},\ \mathbf{b}\in\mathbb{F}_2^m,$$

with all arithmetic over $\mathbb{F}_2$. In the formulation introduced most explicitly in 2025, Stab-QRAM implements the oracle

$$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$

exactly, deterministically, and using only Clifford gates—specifically CNOT and $X$—with $O(\log N)$ logical depth, $O(\log N)$ space for $N=2^n$ data items, and zero $T$-count (Li et al., 30 Sep 2025). Its central premise is that many data-loading tasks do not require arbitrary Boolean lookup; when the target function is affine, the QRAM oracle can avoid the non-Clifford bottleneck that dominates fault-tolerant cost in more general constructions (Li et al., 30 Sep 2025).

1. Position within the QRAM literature

Earlier QRAM literature did not define Stab-QRAM explicitly. The 2023 survey "Quantum Random Access Memory For Dummies" states that it does not discuss “Stabilizer-QRAM” explicitly or in depth, and instead surveys bucket-brigade QRAM, fanout QRAM, flip-flop QRAM (FF-QRAM), qudits-based memory, and approximate PQC-based / EQGAN QRAM (Phalak et al., 2023). The original bucket-brigade qRAM of Giovannetti, Lloyd, and Maccone introduced the now-standard binary-tree access model and emphasized that a qRAM memory call can require only $O(\log N)$ active switches instead of the $N$ used in conventional RAM designs (0708.1879).

The modern significance of Stab-QRAM is best understood against two different baselines. The first is the architectural baseline of bucket-brigade and related tree-based QRAMs, which are designed to support coherent superposition queries over large memories (0708.1879). The second is the complexity-theoretic baseline articulated in "QRAM: A Survey and Critique", which argues that for QRACM any circuit implementing the lookup on a table of size $\mathbb{F}_2$0 needs $\mathbb{F}_2$1 gates in general, and uses this to challenge asymptotically cheap universal circuit QRAM (Jaques et al., 2023). Stab-QRAM does not refute those general concerns. Instead, it changes the problem class: it is intentionally not universal, and its efficiency derives from restricting the supported data to affine Boolean functions (Li et al., 30 Sep 2025).

2. Supported oracle class and Clifford realization

The defining restriction is

$\mathbb{F}_2$2

Here $\mathbb{F}_2$3 is the address, $\mathbb{F}_2$4 computes mod-2 linear combinations of address bits, and $\mathbb{F}_2$5 adds fixed offsets (Li et al., 30 Sep 2025). The corresponding oracle is realized as

$\mathbb{F}_2$6

Operationally, each matrix entry $\mathbb{F}_2$7 means “apply a CNOT from address qubit $\mathbb{F}_2$8 into data qubit $\mathbb{F}_2$9,” and each vector entry $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$0 means “apply an $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$1 to data qubit $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$2” (Li et al., 30 Sep 2025). The oracle is therefore a structured parity-copying network followed by optional bit flips.

This construction places Stab-QRAM entirely inside the Clifford group. The paper states that no Hadamards are needed for the oracle construction itself, and no $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$3-gates or Toffoli gates appear (Li et al., 30 Sep 2025). The restriction is also mathematically sharp: the same work gives a short impossibility argument that a pure Clifford circuit cannot implement a truly universal QRAM oracle, because Clifford circuits map computational basis states to affine transformations over $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$4 (Li et al., 30 Sep 2025). A common misconception is therefore that “all-Clifford QRAM” is merely an implementation choice; in the Stab-QRAM setting it is inseparable from the supported function class.

The target class is nevertheless nontrivial. The paper identifies affine Boolean structure as relevant for discrete dynamical systems, optimization over binary variables, time-series and LFSR-like evolution, linear systems modulo 2, and oracle preparation for QLSAs (Li et al., 30 Sep 2025). Stab-QRAM is thus best viewed as a structured oracle family rather than as a generic memory substitute.

3. Bipartite interaction graphs and optimal logical depth

A central technical result is that the matrix $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$5 induces a bipartite logical interaction graph

$$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$6

whose address-side vertices are $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$7, whose data-side vertices are $$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$8, and whose edges are

$$U_f:\ |\mathbf{x}\rangle|0\rangle^{\otimes m}\mapsto |\mathbf{x}\rangle|f(\mathbf{x})\rangle$$9

The graph is bipartite by construction because every required interaction connects one address qubit to one data qubit, never address-address or data-data (Li et al., 30 Sep 2025).

This graph-theoretic encoding turns circuit scheduling into an edge-coloring problem. A qubit can participate in at most one CNOT at a time, so two CNOTs sharing a qubit cannot occupy the same time layer. In graph language, parallel CNOTs correspond to pairwise non-incident edges. The minimum CNOT depth is therefore the chromatic index $X$0 (Li et al., 30 Sep 2025).

Let

$X$1

This is the maximum number of CNOTs incident on any one qubit. Because $X$2 is bipartite, Kőnig’s edge-coloring theorem gives

$X$3

Accordingly, the CNOT network can be scheduled in exactly $X$4 layers, and this is optimal. The $X$5 gates associated with $X$6 can be applied in one extra parallel layer, so the total logical depth is

$X$7

(Li et al., 30 Sep 2025).

In the worst case, $X$8, hence

$X$9

for $O(\log N)$0 (Li et al., 30 Sep 2025). The significance of this result is not merely asymptotic. The paper’s claim is that the depth bound is proven optimal at the logical scheduling level via graph theory, rather than being a heuristic consequence of a particular decomposition.

4. Resource profile and contrast with general-purpose QRAM

For $O(\log N)$1 memory locations, the circuit uses $O(\log N)$2 address qubits and $O(\log N)$3 data qubits. The total qubit count is therefore $O(\log N)$4, which is $O(\log N)$5 when $O(\log N)$6, the setting emphasized in the paper (Li et al., 30 Sep 2025). Combined with the depth result, this yields the characteristic resource statement: logarithmic space, logarithmic logical depth, exact loading for the supported class (Li et al., 30 Sep 2025).

The practical consequence is the elimination of non-Clifford overhead. Because Stab-QRAM uses only CNOT and $O(\log N)$7, its

$O(\log N)$8

The paper explicitly frames this as avoidance of magic state distillation, with associated reductions in runtime overhead and ancilla overhead, making the architecture especially suited to early fault-tolerant quantum computing platforms (Li et al., 30 Sep 2025).

This differs sharply from the QRAM families emphasized in earlier surveys. Bucket-brigade QRAM uses a binary-tree routing network with qutrit switches and, in circuit form, uses $O(\log N)$9 address qubits, $O(\log N)$0 ancilla qubits for quantum switches, $O(\log N)$1 memory-cell qubits, and one readout qubit, with readout implemented using CNOT and Toffoli gates (Phalak et al., 2023). Fanout QRAM activates $O(\log N)$2 switches even for single-address access, making it more susceptible to decoherence (Phalak et al., 2023). FF-QRAM stores data one by one in superposition but has circuit depth $O(\log N)$3 and width $O(\log N)$4 (Phalak et al., 2023). PQC-based approaches such as EQGAN QRAM and approximate PQC-based QRAM are explicitly approximate and motivated partly by trainability and implementation flexibility rather than exact Clifford structure (Phalak et al., 2023).

A plausible implication is that Stab-QRAM should be compared to those architectures by oracle class as much as by circuit size. It is not a better bucket-brigade, fanout, or FF-QRAM in the universal sense; it is a different point in the design space, where restriction to affine structure buys exactness and zero $O(\log N)$5-count.

5. Application domains

The paper highlights several application domains in which affine Boolean updates arise naturally. In discrete dynamical systems, the evolution

$O(\log N)$6

is exactly of the Stab-QRAM target form. The cited examples include linear feedback shift registers (LFSRs), simplified digital communication models, gene regulatory network abstractions, and certain cryptographic structures (Li et al., 30 Sep 2025).

In optimization, the architecture is positioned for binary optimization problems with affine constraints, including linear systems over $O(\log N)$7, structured binary linear programming, and some forms of combinatorial search (Li et al., 30 Sep 2025). In time-series analysis, the same affine-update structure supports coherent simulation of binary-state dynamics and may assist forecasting or pattern discovery (Li et al., 30 Sep 2025).

The paper also emphasizes Quantum Linear Systems Algorithms (QLSAs), including HHL-type methods, Quantum Interior Point Methods, and Mod2VQLS (Li et al., 30 Sep 2025). In those settings, data-loading and oracle preparation are often bottlenecks. Stab-QRAM is presented as a resource-efficient oracle for structured instances rather than as a general linear-algebra memory substrate.

This application profile is consistent with a broader trend in QRAM research. Algorithmic work on QRAM-limited quantum dynamic programming stresses that quantum speedups can depend sensitively on available qROM or qRAM resources, but that literature treats memory availability abstractly and does not analyze stabilizer memory mechanics directly (Caroppo et al., 2 Apr 2026). Stab-QRAM contributes at the complementary layer: it specifies one concrete structured-oracle regime in which the access primitive itself becomes unusually cheap.

6. Related stabilizer-oriented QRAM directions, limitations, and misconceptions

The term Stab-QRAM is most precise when it denotes the affine-Boolean, all-Clifford architecture of 2025-09-30 (Li et al., 30 Sep 2025). Related QRAM proposals sometimes share stabilizer-oriented features without being the same construction. A distinct example is "Fast and Error-Correctable Quantum RAM", which rewrites the query as

$O(\log N)$8

moves the difficult work into offline preparation of a resource state $O(\log N)$9, and executes the online query using only Clifford operations, Bell measurements, single-qubit Pauli measurements, and Pauli corrections; that scheme reports query depth

$N=2^n$0

and uses $N=2^n$1 ancillas in the online QRAM zone (Cesa et al., 24 Mar 2025). System-level and hardware papers likewise provide stabilizer-adjacent ingredients—such as dual-rail encoding, surface-code-compatible biased-noise strategies, or heralded loss detection—without defining Stab-QRAM explicitly (Xu et al., 2023, Wang et al., 2024).

Two misconceptions recur in this area. First, Stab-QRAM is not a synonym for fault-tolerant QRAM in general. Bucket-brigade, virtual QRAM, network-enabled QRAM, and resource-state QRAM can all interface with fault-tolerance without being the affine-function Stab-QRAM (0708.1879, Xu et al., 2023, Cesa et al., 24 Mar 2025). Second, zero $N=2^n$2-count does not imply arbitrary data loading. The Stab-QRAM paper’s own impossibility argument is that Clifford-only basis-state action is forced to be affine over $N=2^n$3, so universal QRAM lookup is outside the model (Li et al., 30 Sep 2025).

The limitation is therefore structural, not accidental. Stab-QRAM gains its efficiency by restricting the oracle family. This suggests that its natural role is as a specialized subroutine for structured data access, especially in settings where non-Clifford cost dominates the architecture-level budget. Within that role, it supplies an unusually clean result: an exact, all-Clifford, zero-$N=2^n$4 QRAM oracle with optimal logical depth derived from the bipartite interaction graph of the affine map (Li et al., 30 Sep 2025).