RNPU Simulation Results for MNIST and FashionMNIST Classification

Supplementary data from a study on reconfigurable nonlinear-processing units (RNPUs) with capacitive control. The dataset likely contains simulation results demonstrating a tenfold reduction in power consumption and classification accuracies of 95.9% ± 0.2% for MNIST and 86.7% ± 0.2% for FashionMNIST. The data was authored by R. J. C. Cool and last updated on 2026-05-26.

Use Cases

• Benchmarking energy-efficient neuromorphic device designs based on the reported power reduction.
• Training or validating surrogate models for RNPU behavior using the reported classification accuracy metrics.
• Comparing capacitive vs. direct-contact electrode configurations for in-materia computing control.

Strengths

• Reported classification accuracies are quantified with standard deviations (e.g., 95.9% ± 0.2%).
• Demonstrates a specific, quantified tenfold reduction in power consumption compared to a prior implementation.
• Data is associated with a peer-reviewed research concept published in Nature.

Limitations

• The dataset is a 899.0 KB DOCX file, which suggests limited scope and may contain primarily textual or tabular summary results rather than raw simulation data.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale model training.

Provenance

Source

figshare

Collection Method

Likely contains results from experimental measurements and device simulations based on a surrogate model.

Freshness

Last updated 2026-05-26 06:01:59; freshness should be verified.