Every product company eventually faces the question: is our differentiator big enough to justify custom silicon? The answer has changed over the past decade — design starts at mature nodes have become affordable enough that ASICs now pencil out at volumes that once looked absurd — but the financial logic has not. Here is how the decision is actually made.
The Cost Structure, Honestly Stated
- NRE: a full-custom design at 28 nm runs roughly $5–20 M including IP, verification, and mask set; mature nodes (180 nm–90 nm) can land near $1–3 M. The mask set alone at advanced nodes exceeds $5 M.
- Unit cost: die cost follows wafer pricing divided by yield; at volume, a custom part routinely undercuts an equivalent catalog IC by 30–60 % — but only at volume.
- Break-even volume: the rule of thumb is that custom silicon pays for itself somewhere between 100 k and 1 M units per year depending on complexity and node. Below that, you are buying prestige with margin.
The Volume Ladder: Options Between "Buy" and "Full Custom"
| Approach | NRE | Break-Even Volume | Best For |
|---|---|---|---|
| Off-the-shelf + firmware | ~0 | — | Most products; differentiation in software |
| FPGA / programmable SoC | Low–moderate | — | Volumes < 50 k/yr, evolving algorithms |
| Structured / platform ASIC | $0.5–3 M | ~50–200 k/yr | Mid volumes with power/cost pressure |
| Full-custom ASIC | $5 M+ | > 200 k–1 M/yr | Flagship products, extreme integration |
The hybrid path — off-the-shelf analog and mixed-signal building blocks plus an ASIC that integrates the digital glue — is where most mid-volume projects land. It buys the integration and cost benefits where they matter while leaving the hard analog IP to companies that have already solved it.
What Buyers Underestimate
- Time and iteration risk: 18–30 months from spec to production silicon for a full custom part, and silicon respins are neither rare nor cheap. A design freeze that arrives too early locks in yesterday's requirements.
- Software and ecosystem: a custom part means custom drivers, toolchains, and documentation — ongoing headcount, not a one-time cost.
- Single-sourcing by definition: nobody second-sources your ASIC. The supply continuity argument that favors catalog parts flips entirely; mitigation requires mask ownership, test program escrow, and fab agreements — negotiate these before tape-out.
- Verification is the cost driver: most overruns are verification overruns, not logic design. Budget realistically.
The Cases Where It Clearly Makes Sense
- Volume is proven and stable: the product line's forecast rests on shipped history, not a business plan.
- The differentiator is genuinely architectural: power, latency, or integration that catalog parts cannot reach.
- BOM economics dominate: the part sits in the cost-critical path and volumes amortize NRE quickly.
- Supply control is strategic: some companies accept custom silicon partly to own the supply chain — a real benefit when negotiated deliberately.
For everyone else, today's off-the-shelf ICs are extraordinarily capable, and the fastest route to market usually wins. JTDZ Tech supports both paths: catalog ICs from every major manufacturer, and sourcing guidance for custom-silicon programs. Tell us where your product sits on the volume ladder and we will help you cost the realistic options.
