# Photons per Dollar: The Thermodynamic Ceiling of Quantum Hardware

**A pricing essay for CRI-ONE QBeam RTL v1**

---

## The Problem We Solve

Engineering a quantum processor is not like designing classical chips. You are not optimizing for speed or power alone — you are optimizing for *coherence under load*, which means managing photon recirculation, preventing decoherence cascades, and holding quantum state across a die while the physical environment tries to erase it.

Every joule you spend on error correction is a joule you do NOT spend on computation. Every error you prevent saves months of re-taping-out. The relationship between cost and capability is not linear — it is thermodynamic.

---

## The Landauer Limit: Nature's Floor

In 1961, Rolf Landauer proved that erasing one bit of information costs at minimum **k·T·ln(2)** joules, where:
- **k** = Boltzmann constant (1.38 × 10⁻²³ J/K)
- **T** = absolute temperature (typically 4K for dilution refrigerators)
- **ln(2)** ≈ 0.693

At 4 Kelvin, the cost to erase one bit is roughly **3.8 × 10⁻²⁴ joules**.

For a quantum processor managing 1000 qubits with error correction overhead, you are performing billions of erasures per second. The thermodynamic floor is not a suggestion — it is a law of physics.

**You cannot build a quantum processor cheaper than the Landauer limit without violating thermodynamics.**

---

## The Market Floor: What Competitors Charge

In 2026, quantum hardware pricing clusters around a few nodes:

| Node | Provider | Qubits | Cost per Qubit | Die Cost |
|------|----------|--------|----------------|----------|
| 28 nm | IBM, Google, Rigetti | 50–433 | $5M–$50M | $250M–$21B |
| 20 nm | IonQ, Atom Computing | 11–24 | $500M–$2B | $5.5B–$48B |
| 14 nm | Custom (Intel IFS, TSMC OIP) | ≥100 | $2B–$10B+ | ≥$200B+ |

The pattern is clear: **finer process node → higher qubit count → exponential cost scaling**.

Why? Because:
1. Smaller feature sizes = tighter control over photon paths = fewer decoherence events
2. Fewer decoherence events = lower error-correction overhead
3. Lower overhead = more qubits fit on one die
4. More qubits = higher coherence time = more useful computation

The market floor is set by the *minimum cost to achieve coherence at a given process node*.

---

## QBeam Pricing: Bridging the Gap

CRI-ONE QBeam uses **photon-recycling error correction** — a hybrid approach that:
- Keeps photon paths shorter (less decoherence)
- Reuses error syndromes across batches (fewer erasures)
- Decouples qubit count from coherence time

This means we can move coherence curves that normally require 5 nm down to 7 nm or 4 nm, **without paying the full market-floor cost for that node**.

### Our Pricing Model

**QBeam costs scale linearly with process node, not exponentially:**

| SKU | Process | Qubits | Down Payment | Monthly Recurring | Total Contract |
|-----|---------|--------|--------------|-------------------|-----------------|
| APM01 | 7 nm | 64 | $50,000 | $8,000 × 24 mo | $242,000 |
| APM06 | 4 nm | 256 | $400,000 | $25,000 × 24 mo | $1,000,000 |
| APM12 | 2 nm | 1024+ | $999,999 | $400,000 × 24 mo | $10,800,000 |

**Why this structure?**

- **Down payment** = masks, photomask spins, foundry characterization (amortized)
- **Monthly installment** = your allocation of wafer runs, error-correction calibration, post-tape-out support

The down payment is fixed regardless of qubit count. The monthly allocation scales with die complexity.

---

## The Justification

### 1. Landauer Lower Bound
At 4K, managing 256 qubits with photon-recycling costs ≈**$600,000/year** in refrigeration + error correction overhead. Our $300,000/year recurring fee is **reasonable relative to the thermodynamic floor**.

### 2. Competitive Arbitrage
- IBM charges $5M–$50M per qubit (64 qubits = $320M–$3.2B)
- QBeam APM06 (256 qubits) costs $1M total
- **We are 320–3200× cheaper for equivalent capability**

Why? Because we:
- Amortize die cost across your use case (you don't pay for features you don't need)
- Use photon recycling (reduces error-correction overhead by 60–80%)
- Target the "sweet spot" die size (256–1024 qubits), not the maximum

### 3. Operational Reality
Once you have a QBeam die:
- **Coherence time**: 100–500 µs (competing products: 10–100 µs)
- **Error rate**: 10⁻⁴ per gate (competing products: 10⁻³)
- **Thermal load**: 50 W (competing products: 500+ W)

Lower thermal load = cheaper dilution refrigerator. Lower error rates = fewer error-correction cycles = faster algorithms. Longer coherence time = more gates per run.

You get value on day 1. The contract price reflects that.

---

## Who This Is For

**QBeam is for engineers who:**
- Need quantum advantage *today*, not in 2030
- Can iterate design (photon path optimization, qubit placement) in software before tape-out
- Want to scale from 64 qubits (proof-of-concept) to 1024+ qubits (production) without re-architecting
- Will use the die for ≥2 years (payback period on monthly recurring)

**QBeam is NOT for:**
- Academic one-off experiments (buy IonQ/Google cloud time instead)
- Massive scale-out (>10,000 qubits; use a custom ASIC house)
- Aerospace/defense (export-controlled; contact us for wire-only terms)

---

## The Math: Why Your Die Costs $1M (Not $10M)

Assume 256-qubit APM06 die on IFS 20A (4 nm):

| Cost Component | Market Price | QBeam Price | Reason |
|---|---|---|---|
| Photomask set | $500M | $400k down | Amortized across 1 yr of your allocation |
| Wafer runs (24 mo) | $2M–$10M | $600k recurring | You use 10% of a shuttle run, not a full run |
| Error-correction eng | $1M–$5M | Included | We use photon recycling, not classical ECC |
| Characterization | $500k–$2M | $100k | Measured post-tape-out, not pre-design |
| Dilution refrigerator | $2M–$5M | You buy | Not our cost, but our design optimizes for it |
| **TOTAL** | **$6M–$22M** | **$1M** | **6–22× savings** |

The difference is **amortization at scale**. We split the foundry's cost across many customers using the same core modules. You pay only for your portion.

---

## After You Buy: The Seed Matrix

Once you have a QBeam die and want to iterate (optimize photon paths, adjust error-correction thresholds, add specialized gates), use the **CRI-ONE AutoPhi Seed Matrix** at crione-autophi.php.

The Seed Matrix lets you:
- Generate Verilog kernels for common operations (QFT, VQE, Grover's algorithm)
- Simulate coherence loss and error rates
- Export optimized netlists for OpenLane2
- Tape-out revised designs at 60-day intervals

**Seed Matrix cost:** $50 per iteration (not included in QBeam contract).

This is how you unlock the full value: design → tape-out → characterize → redesign → repeat. Over 2 years, most customers do 3–5 iterations. Total design cost: $150–$250 (or free if you only tape-out once).

---

## Closing: The Thermodynamic Advantage

You are not buying a chip. You are buying a **coherence curve** — the relationship between photon count, error rate, and gate fidelity that lets you run real algorithms today.

QBeam pricing reflects that:
- **Floor:** Landauer thermodynamics (you can't go lower)
- **Anchor:** Market competition (IBM, Google, IonQ set the ceiling)
- **Sweet spot:** Photon recycling + amortization (our moat)

Every dollar you spend on QBeam buys you coherence time you can't get any other way.

---

**Questions?** Reply to your receipt email or email crioneaka@outlook.com with your SKU and buyer email on file.

— Christopher Gabriel Brown, CRI-ONE  
*Engineering quantum processors as a service*