The Real Problem

Computing's wall isn't speed anymore. It's energy and heat.

• Every logic operation dissipates power — at scale, cooling rivals the compute itself
• The frontier isn't a faster transistor; it's switching without the thermal penalty
• Yet the industry keeps iterating on the same handful of materials

The Bet

Some materials switch between conducting and insulating by moving ions — electrochemical potential — instead of brute thermal energy.

This isn't theory. It already exists:

• Memristors / ReRAM — resistance set by ion migration
• ECRAM — electrochemical switching built for low-energy neuromorphic logic
• Mott transistors — voltage-driven metal↔insulator flips, no heat

The open question: has anyone systematically searched for the lowest-energy switch — or is everyone betting on the same few?

The Insight

Treat material discovery as a screening problem, not a wait-for-luck problem.

Rank candidate material systems by ion-migration barrier and switching energy — computationally, before anyone touches a fab. Discard the heat-heavy. Surface the unconventional.

What's Actually Missing

The data isn't absent — it's scattered. NIST, the Materials Project, thousands of papers, countless lab notebooks.

What doesn't exist: a unified, audited, provenance-tracked shortlist of athermal switching candidates, ranked by how cheaply they switch.

That's what we build.

The Method

An AI research fleet runs the screen end-to-end:

• Mine — pull properties from open databases and the literature
• Model — DFT / NEB migration-barrier and switching-energy screening per system
• Rank — order candidates by switching energetics; flag anomalies
• Trace — every value tied to its source; every dead-end logged

Material Systems, Not Bare Elements

Switching lives in compounds, dopants, and interfaces — oxides, filament stacks — not isolated elements.

• We search the system space, not the periodic table in a vacuum
• Transition-metal oxides stay in scope — today's frontier, not legacy
• The unit of discovery is a switch, not a square on a chart

Integrity by Design

The screen is only worth what its audit trail is worth.

• Pre-registration — criteria locked before the run
• Multi-layer audit — format, physics-plausibility, anomaly review
• Failures kept — dead-ends logged with reasoning, never deleted
• Full provenance — every candidate reproducible from the log

Swap the model. Keep the audited brain.

Phase 1 — The Deliverable

A public, ranked, fully-audited shortlist of athermal switching material candidates worth fabricating.

The dataset is the product — open, reproducible, built to be cited.

This is the part we can deliver now, with the agent fleet already built.

Phase 2 — Prove the Switch

Take the top candidates from the shortlist and prototype athermal logic — with a fab, university, or industry partner.

• Material → working chip is the entire industry's stack — we don't fab in a garage
• Phase 1 is the evidence that earns the partnership
• The shortlist de-risks the bet before a dollar of silicon is spent

Why Now

• AI's compute demand is colliding with a hard energy wall
• Neuromorphic and in-memory computing are gaining real momentum
• Owning your compute stack — not renting it — is now a strategic priority
• AI fleets can finally screen the materials space at a scale humans can't

What We Need

• Compute — GPU/CPU capacity for the DFT screening runs
• A materials-science collaborator — to validate methodology and ground the physics
• A Phase-2 partner — a fab or lab to prototype the winners
• Phase-1 funding — sized to a screening program, not a factory

Two Phases, Two Asks

Fund the search first. Earn the fab second.

Find the cold switch.
Then build it.

A systematic, audited search for computing without the heat.

Project 118 · Developum AI Engine · 2026