Commercial Viability Assessment: Hosted Co-Design Platform¶
Executive Summary¶
This document evaluates whether OpenLithoHub should pursue a commercial hosted co-design layer on top of the open-source core. It is structured as a demand-validation probe: customer profiles, value propositions, pricing models, and a go/no-go framework.
Target Customer Profiles¶
Profile A: Fabless Semiconductor Startup¶
- Size: 10-50 engineers
- Pain point: No in-house OPC/ILT team, cannot justify a Synopsys CATS or Cadence Litho Analyzer license ($500K+/year)
- Need: Push-button mask optimization with reasonable quality
- Budget: $5K-20K/month for tools
- Decision maker: CAD/EDA team lead or VP Engineering
Profile B: University Research Lab¶
- Size: 3-10 researchers
- Pain point: Need to benchmark novel ILT algorithms against baselines but lack lithography simulator infrastructure
- Need: Standardized benchmarking environment with leaderboard
- Budget: $500-2K/month (grant-funded)
- Decision maker: PI or postdoc leading the project
Profile C: Established IDM / Foundry R&D¶
- Size: 100+ engineers
- Pain point: Internal tooling exists but is slow to iterate; need rapid prototyping for new process node exploration
- Need: Co-design sandbox that integrates with internal flows via API
- Budget: $20K-100K/month for targeted tools
- Decision maker: DFM group manager or lithography R&D director
Profile D: Multiphysics Design Consultant¶
- Size: 1-5 engineers
- Pain point: Client projects span lithography + CFD + thermal; no single tool covers all
- Need: Co-design platform linking DiffCFD + OpenLithoHub + DiffNano
- Budget: $1K-5K/month
- Decision maker: Principal engineer / company owner
Value Proposition: Hosted vs. Self-Hosted¶
| Dimension | Self-Hosted (OSS) | Hosted (Commercial) |
|---|---|---|
| Setup time | Hours-days (GPU drivers, deps) | Minutes (web UI / API key) |
| GPU hardware | Customer must provision | Included |
| Model selection | Manual config | Auto-tuned per process node |
| Multi-GPU scaling | Manual multiproc setup | Automatic load balancing |
| Monitoring | Roll your own | Dashboard with alerts |
| Support | Community only | SLA-backed, priority queue |
| Data privacy | Full control (on-prem) | Cloud-hosted (NDA available) |
| Cost | Hardware + engineer time | Subscription fee |
Key Differentiator¶
The hosted layer must justify its cost against the self-hosted option. The primary value is time-to-first-result: an engineer at a fabless startup should get a optimized mask within 15 minutes of signing up, versus 2-3 days of setting up self-hosted infrastructure.
Pricing Model Options¶
Option 1: Per-Optimization-Run¶
- $0.50-5.00 per optimization run (based on problem size)
- Good for: sporadic users, research labs
- Risk: unpredictable revenue; heavy users churn to self-hosted
- Unit economics: GPU time costs ~$1-3/hr on cloud; typical run = 1-10 min
Option 2: Monthly Subscription (Tiered)¶
| Tier | Price | Included | Target |
|---|---|---|---|
| Free | $0 | 10 runs/month, community support | Researchers, evaluation |
| Pro | $500/month | 500 runs/month, email support | Small teams |
| Team | $3,000/month | 5,000 runs/month, API access, priority support | Mid-size companies |
| Enterprise | Custom | Unlimited, on-prem option, SLA | IDMs, foundries |
- Good for: predictable revenue, budget planning
- Risk: underuse (customers pay but don't use) or overuse
Option 3: GPU-Time Metered¶
- $2-4/hour of GPU compute consumed
- Good for: heavy users who want cost proportional to usage
- Risk: bill shock; hard to budget
Recommendation¶
Start with Option 2 (subscription) because: 1. Predictable revenue enables capacity planning 2. Free tier drives adoption and benchmarking data 3. Enterprise tier captures high-value customers 4. Simpler billing than metered GPU time
Customer Interview Questions¶
For Fabless Startups (Profile A)¶
- What is your current OPC/ILT workflow? (manual, commercial tool, none?)
- How many mask optimizations do you run per month?
- What is your tolerance for optimization quality vs. cost? (90% of commercial quality at 10% of cost?)
- Would you trust a cloud-hosted service with your design data under NDA?
- What is your target process node? (28nm, 14nm, 7nm, 5nm?)
- How long does your current mask optimization take per run?
For University Labs (Profile B)¶
- What baselines do you currently compare against?
- Do you need GPU access, or do you have your own?
- Is reproducibility across papers important to you?
- Would a public leaderboard drive adoption of your research?
- What is your grant cycle for tool budget?
For IDMs (Profile C)¶
- Where does your internal tooling fall short?
- What would a co-design sandbox need to integrate with? (GDS/OASIS in/out? KLayout bridge? Calibre DRC hooks?)
- Is a hosted API acceptable, or do you require on-prem deployment?
- What is your evaluation process and timeline for new tools?
- What existing commercial tools would this displace or complement?
For Consultants (Profile D)¶
- What domains do your projects span? (litho + CFD? thermal + optics?)
- Do clients require specific output formats? (GDS, OASIS, CIF?)
- How do you currently handle multiphysics optimization?
- Would API-based access fit your workflow?
Decision Criteria¶
Go Signals (Invest in Commercial Layer)¶
- [ ] 5+ potential customers interviewed, 3+ express willingness to pay
- [ ] At least one customer commits to a pilot ($1K+ MRR)
- [ ] Self-hosted setup takes >4 hours for a new user (it currently does)
- [ ] Competitor pricing leaves a gap we can fill at >70% margin
- [ ] OpenLithoHub benchmarking data has commercial value (leaderboard, process node coverage)
No-Go Signals (Stay Open-Source Only)¶
- [ ] All interviewed customers prefer self-hosted for data privacy
- [ ] No one willing to pay >$500/month
- [ ] Internal tooling at target customers already covers our use case
- [ ] Engineering effort to build hosted layer > 6 months of team time
- [ ] GPU cloud costs make unit economics negative at viable price points
Key Metrics to Track¶
| Metric | Target (6 months) | Measurement |
|---|---|---|
| Interview count | 20+ potential customers | CRM log |
| Conversion rate (trial -> paid) | >10% | Billing system |
| Monthly active users (free tier) | 500+ | Analytics |
| Paying customers | 10+ | Billing system |
| MRR | $10K+ | Finance |
| Time-to-first-result | <15 min | Product analytics |
| NPS | >40 | Survey |
Go/No-Go Decision Framework¶
Interview Results
|
+-----------+-----------+
| |
Positive signal Negative signal
(3+ willing to pay) (none willing to pay)
| |
Build MVP Stay OSS-only
(3 months) (no commercial layer)
|
Pilot with 3 customers
(3 months)
|
+----+----+
| |
Pilot Pilot
succeeds fails
(>3 paying) (<3 paying)
| |
Full launch Sunset pilot
(hire sales) (refocus on OSS)
Phase Gate Schedule¶
- Month 1-2: Customer interviews (20+), pricing research
- Month 3: Go/No-Go decision point
- Month 3-5: MVP build (if Go) -- web UI, API, billing, GPU autoscaling
- Month 6-8: Pilot with 3-5 customers
- Month 9: Scale decision -- continue, pivot, or sunset
Minimum Viable Product Scope¶
If we proceed, the MVP should include:
- Web UI for mask upload + optimization
- REST API for programmatic access
- 3 model options (Neural-ILT, GAN-OPC, Surrogate-ILT)
- Basic billing (Stripe integration)
- Results dashboard with quality metrics
- GDS/OASIS download
Estimated Investment¶
| Item | Cost (3 months) |
|---|---|
| 2 engineers (full-stack + ML infra) | $60K |
| GPU cloud (development + pilot) | $5K |
| Stripe/billing setup | $2K |
| Domain, SSL, monitoring | $1K |
| Total | $68K |
Break-even at $10K MRR = ~7 months after launch (assuming steady 70% gross margin).