Coupled Quantum Reservoir Features for Multi-Chemistry Battery State-of-Health Estimation | SAKTHI PRABHU GUNASEKAR, PRASANNA KUMAR RANGARAJAN | ResearchPod

Transcript: Coupled Quantum Reservoir Features for Multi-Chemistry Battery State-of-Health Estimation

Alex: Welcome to another episode of ResearchPod. Today we're looking at a paper on battery state-of-health estimation that challenges a common assumption in quantum machine learning.

Sam: The paper asks a pointed question: do we actually need recurrent quantum memory to track battery degradation? The authors propose what they call a Coupled Quantum Reservoir Framework—CQRF—which uses a fixed, non-recurrent circuit to extract features. And they find it outperforms recurrent designs.

Alex: So the claim is that the "memory" we usually build into quantum reservoirs is actually a liability for modeling sequential battery data?

Sam: That's the core argument. Battery management systems need to estimate state-of-health across diverse chemistries and charge histories. Recurrent quantum reservoirs try to do that by feeding state information back into the circuit—essentially accumulating a kind of degradation memory. But that feedback loop creates two practical problems: warmup periods where the model needs several cycles before its internal state is meaningful, and state-propagation errors that compound over time. The authors argue a fixed, static context sidesteps both.

Alex: Walk me through the architecture. What does "fixed and static" actually mean in circuit terms?

Sam: The circuit has ten qubits split into two registers. Four qubits form what they call the auxiliary register—this is the fixed lens. It's initialized once and never updated between cycles. The other six are the processor register, which receives the actual battery measurements. The key move is cross-register entanglement: by coupling these two registers, the auxiliary qubits condition how the processor encodes the input, creating a richer feature representation than you'd get from the processor alone. Crucially, there are no trainable quantum parameters. The entire learning burden falls on a classical readout layer sitting on top.

Alex: So the quantum circuit is doing feature extraction, and a classical layer is doing the regression.

Sam: Exactly. Which is why the parameter count is so low—around 125 trainable parameters total. The quantum part is fixed; only the readout weights are optimized. That's the source of both its efficiency and its constraint.

Alex: And that's where the tension is. If the circuit isn't evolving, how does it capture the history of a battery's degradation?

Sam: That's the right place to push. The honest answer is: it doesn't, not explicitly. What it does instead is generate a high-dimensional feature space from each cycle's inputs—voltage, current, temperature—conditioned by the auxiliary register. The bet is that those features are expressive enough that a linear readout can recover the health estimate without needing an explicit memory of prior cycles. And across the datasets they tested, that bet pays off at the macro level. The recurrent baseline wins on a few specific subsets, but CQRF is more consistent overall.

Coupled Quantum Reservoir Features for Multi-Chemistry Battery State-of-Health Estimation

SAKTHI PRABHU GUNASEKAR, PRASANNA KUMAR RANGARAJAN

Jan 1, 2026·5:41·doi·Original Paper

Coupled Quantum Reservoir Features (CQRF)Battery State-of-Health (SOH) EstimationCompact Bottleneck Readout (CBR)NISQ Quantum HardwareCross-dataset GeneralizationQuantum Reservoir Computing (QRC)

Listen

6 min

Briefing

Research Question

Battery state-of-health (SOH) estimation is critical for electric vehicle safety and longevity. While recurrent neural networks and recurrent quantum reservoirs are often used to model the sequential nature of battery degradation, they face challenges with short-cycle datasets and high parameter requirements. This paper investigates whether a non-recurrent, coupled quantum reservoir can capture degradation signatures as effectively as recurrent quantum memory, while maintaining parameter efficiency.

Approach

The authors propose Coupled Quantum Reservoir Features (CQRF), a fixed ten-qubit quantum reservoir. The architecture consists of a 4-qubit auxiliary context register and a 6-qubit processor register. Unlike recurrent models, the auxiliary register is reset for every cycle, and the circuit is never trained. Instead, the model uses a Compact Bottleneck Readout (CBR) with only 125 trainable parameters to map the 57-dimensional quantum observables to SOH estimates. The authors compare CQRF against a recurrent Quantum Recurrent Reservoir (QRR) baseline, ensuring both models share identical datasets, features, and parameter counts.

Main Findings

CQRF+CBR achieved a macro mean absolute percentage error (MAPE) of 0.0172 across four diverse battery datasets (NCM, LFP, and NCA/NCM-blend). In a controlled comparison, CQRF outperformed the recurrent QRR baseline on all four dataset-level means, demonstrating that recurrent quantum memory is not strictly necessary for accurate SOH estimation in this context. Furthermore, CQRF requires significantly fewer parameters than physics-informed neural networks (QPINN), offering a 184-fold reduction in trainable complexity. Hardware audits on IBM's 156-qubit processor confirmed the feasibility of the design, though they also highlighted current limitations in simulator-hardware feature correlation.

Why It Matters

This work challenges the assumption that temporal recurrence is essential for quantum-based sequential modeling in battery health. By demonstrating that a fixed, coupled quantum reservoir can outperform recurrent alternatives, the authors provide a pathway for deploying parameter-efficient quantum models on near-term noisy hardware. The study also establishes a rigorous protocol for comparing quantum reservoir designs, which is vital for the development of practical quantum machine learning applications.

Key Terms and Definitions

Coupled Quantum Reservoir Features (CQRF) — A fixed, non-recurrent quantum reservoir architecture that uses a two-register design to extract features from battery cycles independently.
Compact Bottleneck Readout (CBR) — A small, 125-parameter neural network used to map high-dimensional quantum observables to SOH predictions.
Quantum Recurrent Reservoir (QRR) — A baseline architecture that propagates degradation history across cycles using a memory register updated by Z-basis measurement feedback.
State-of-Health (SOH) — The ratio of a battery's current discharge capacity to its initial capacity, serving as the target metric for degradation estimation.
NISQ Hardware — Noisy Intermediate-Scale Quantum devices, which are currently limited by noise and decoherence but used here to audit the feasibility of the proposed quantum circuits.

Coupled Quantum Reservoir Features for Multi-Chemistry Battery State-of-Health Estimation: Research Paper Podcast