Publications

Research publications

Preprints and papers released under the XSYDA Research banner, with an emphasis on reproducibility and honest reporting.

Preprint 001

Beyond Synthetic Benchmarks: Reproducible Evaluation of Engineering-Intelligence Systems for Semiconductor Design

A reproducible, anti-circular, provenance-aware framework for evaluating engineering-intelligence systems, applied to real open-source RTL with maintainer-derived ground truth. The paper introduces XRecall, a provenance-anchored cross-project engineering-memory architecture, and reports a candid account of findings — including a documented synthetic-to-real generalization gap — making no superiority claim over existing tools.

Author
Srinivasan Subramanian
Type
Research Preprint
Version
1.0
Published
May 2026
Download PDF
Preprint 002

The AI Demand Shock: A Falsification Test of AI-Era Demand Compression

An evidence-based investigation into one of the most widely repeated assumptions in modern retail and supply-chain planning: that artificial intelligence has fundamentally compressed consumer demand cycles. Using a falsification framework and lifecycle analysis across consumer products, the study examines whether AI-era demand exhibits measurably faster emergence, peak formation and decay than comparable pre-AI trends. Across matched categories the evidence does not support the compression hypothesis — lifecycle compression tracks product type, not era.

Author
Srinivasan Subramanian
Type
Research Preprint
Version
1.0
Published
June 2026
Download PDF
Preprint 003

Beyond Silicon: A First-Principles Framework for Materials Required by Future Intelligence Systems

A first-principles materials-science framework for the substrate of future intelligence systems (2035–2050). Rather than seeking a single silicon successor, the paper argues — via the XSYDA Data-Movement Principle and a five-tier Unified Energy Hierarchy — that the decisive gains come from materials that move data with less energy and compute where the data already resides. It evaluates nine material categories across fifteen criteria spanning charge transport, advanced doping and junction engineering, wide-bandgap power, thermal management, thermionic cooling, photonics and compute-in-memory, and concludes with a silicon-anchored, function-partitioned heterogeneous architecture.

Author
Srinivasan Subramanian
Type
Research Preprint
Version
1.0
Published
June 2026
Download PDF