Why a formula?
So you can check our work. Every score on this site comes from the same recipe below. Anyone with the sources can reproduce it, and argue with our inputs.
1. Readiness (0-100): how close to making money
| Criterion | Weight | How we score it (0-5) |
|---|---|---|
| Evidence level | 55% | How far along the proof is (TRL-style ladder). |
| Manufacturability | 20% | 0 = exotic materials no fab can make; 5 = runs on today's standard lines. |
| Supply chain | 10% | 0 = tools/materials don't exist; 5 = mature suppliers ready. |
| Corporate backing | 10% | 0 = academic only; 2 = funded startup; 3 = big-company R&D; 5 = several big firms committing capex. |
| Independent groups | 5% | 1 = one lab; 5 = industry-wide effort. |
Evidence ladder (the biggest weight)
- Paper / simulation only → 0% of the evidence weight · Stage 1 · years-to-market 6-10
- Lab demo → 10% of the evidence weight · Stage 1 · years-to-market 4-8
- Working prototype chip → 25% of the evidence weight · Stage 2 · years-to-market 3-6
- Customer samples / qualification → 45% of the evidence weight · Stage 2 · years-to-market 2-4
- Volume production or orders → 75% of the evidence weight · Stage 3 · years-to-market 0-2
- Revenue at scale → 100% of the evidence weight · Stage 4 · years-to-market now
2. Other scores
Set by evidence only: 0-1 → Stage 1 Paper, 2-3 → Stage 2 Prototype, 4 → Stage 3 Factories & orders, 5 → Stage 4 Revenue.
From the evidence ladder. Early items (evidence 0-2) get +2 years if manufacturability is 0-1, or −2 years if it already runs on standard lines (4-5).
Hype = 10 × claim boldness (0-5; "1000× faster" = 5) + 10 × attention (0-5: press, funding, famous backers). Gap = Hype − Readiness: up to 10 = reality catching up, up to 35 = some hype, more = mostly hype.
Average of crowding (how many patent holders, 0-5) and litigation history (0-5): 3.5+ High, 2+ Medium, else Low. Not legal advice.
Score = 0.4 × average readiness of linked research + 0.4 × 20 × exposure (share of the business riding on it, 0-5) + 0.2 × 20 × execution (track record, balance sheet, 0-5). 65+ High, 45-64 Medium, under 45 Speculative. Not directly investable, or exposure 0-1 → Watch. Execution 0-1 (e.g. pre-profit, funding risk) caps it at Speculative.
3. Worked example: Gluing chips together copper-to-copper (hybrid bonding)
| Criterion | Sub-score | Points |
|---|---|---|
| Evidence level | 4/5 | 41.2 |
| Manufacturability | 4/5 | 16.0 |
| Supply chain | 4/5 | 8.0 |
| Corporate backing | 5/5 | 10.0 |
| Independent groups | 5/5 | 5.0 |
| Readiness | 80 | |
4. Re-score on 2026-10-29: what changed
Our first issue used expert judgment directly. Re-running every item through this formula moved these numbers. Stage changes: Neuromorphic S2→S1.
| Item | Readiness | Hype | Years | Stage |
|---|---|---|---|---|
| Computing with light instead of electricity (optical | 25 → 31 | 80 → 80 | 4-6 → 3-6 | S2 → S2 |
| A light-based 'highway' between AI chip and memory | 20 → 32 | 75 → 80 | 4-7 → 4-8 | S1 → S1 |
| Storing many values in one memory spot (superposed a | 15 → 19 | 40 → 40 | 5-8 → 6-10 | S1 → S1 |
| Making noisy analog AI chips accurate (weight splitt | 30 → 35 | 30 → 30 | 3-5 → 4-8 | S1 → S1 |
| A light switch sending 200 billion bits per second p | 35 → 36 | 45 → 40 | 3-5 → 2-6 | S1 → S1 |
| Stacking atom-thin transistors on top of each other | 10 → 16 | 55 → 50 | 8+ → 6-10 | S1 → S1 |
| Gluing chips together copper-to-copper (hybrid bondi | 65 → 80 | 50 → 50 | 1-3 → 0-2 | S3 → S3 |
| Glass replacing plastic under AI chips (glass substr | 40 → 52 | 65 → 70 | 2-4 → 2-4 | S2 → S2 |
| Brain-like chip that can learn on the device (neurom | 25 → 28 | 60 → 60 | 4-6 → 4-8 | S2 → S1 |
| One transistor that can be logic, memory or neuron | 10 → 16 | 35 → 30 | 7+ → 6-10 | S1 → S1 |
| Light-powered brain chip: 1,176x faster on vision | 15 → 16 | 70 → 70 | 5-8 → 6-10 | S1 → S1 |
| Silicon photonics reaches high-volume factories | 75 → 84 | 25 → 30 | 0-2 → 0-2 | S3 → S3 |
Conviction changes
| Company | Conviction | Score |
|---|---|---|
| Ibiden | Speculative → Medium | 61 |
| Tower Semiconductor | Speculative → Medium | 49 |
| Corning | Speculative → Medium | 53 |
| BrainChip | Watch → Speculative | 55 |
| Besi | Medium → High | 88 |
| Kaynes Technology | Speculative → Watch | 34 |
| CG Power | Speculative → Watch | 52 |
| Adeia | Speculative → Medium | 64 |
| STMicroelectronics | Medium → High | 66 |
| TSMC | High → Medium | 57 |
| AMD | Medium → Watch | 39 |
5. Data sources & update cadence
- Papers: Nature-family journals, arXiv, AIP and conference proceedings (IEDM, ISSCC).
- Patents: Google Patents and patent-analytics sites.
- Company facts: SEC filings, investor-relations releases, earnings reports.
- Counter-evidence: trade press, reviews and company filings, always linked in each "What could go wrong" box.
- Cadence: one issue per day when there is news. Each item keeps first-seen, last-updated and history, and every call is logged on the Track record.
6. Limitations (please read)
- Sub-scores are human judgment. Two careful analysts could differ by ±1 on any criterion (about ±10 readiness points).
- Weights are our choice, not a statistically validated model. Any change to them will be published here.
- Press releases overstate; papers under-report failures. A missing bear source does not mean low risk.
- We have no access to private company data, customer contracts or patent claim charts.
- Conviction is about the research-to-revenue story, not share price or valuation.
⚠️ Not investment, legal or financial advice. Do your own research and consult a registered adviser.