ASIC Development Costs: Why So High & How to Estimate

ASIC (Application-Specific Integrated Circuit) development is expensive because you are designing and manufacturing a custom silicon chip from scratch, with non-recurring engineering (NRE) costs typically ranging from $1 million to over $50 million depending on the process node and complexity. To estimate costs, you must account for architecture design, verification, physical design, mask sets, prototyping, and packaging, each of which scales with the chip’s size and the manufacturing technology. This article breaks down each cost driver and gives you a practical framework for building your own estimate.

Why ASIC Development Is So Expensive

Unlike buying an off-the-shelf microcontroller or FPGA, an ASIC is designed for one specific purpose. Every transistor, memory block, and I/O pad is custom. The high cost comes from two main areas: engineering labor (salaries for a large team over 1-3 years) and manufacturing setup (mask sets, wafers, packaging). The engineering effort is dominated by verification—proving the design works before spending millions on fabrication. The manufacturing cost is dominated by mask sets, which can cost $1-5 million at 28nm and over $10 million at 7nm or below.

Another factor is the tiny volume of chips produced relative to the design effort. If you only need 10,000 chips, the NRE cost is amortized over just 10,000 units, making each chip extremely expensive. This is why ASICs only make sense for high-volume products (millions of units) or applications where performance, power, or size requirements cannot be met by any other solution.

Key Cost Components

To estimate the total cost, break it down into these major categories:

• Architecture and RTL design: Writing the hardware description language (VHDL/Verilog) that defines the chip’s logic. This includes microarchitecture decisions, IP selection, and system integration.
• Verification: Creating testbenches, simulations, and formal proofs to ensure the design is bug-free. Verification typically consumes 50-70% of the total engineering effort.
• Physical design: Converting the RTL into a physical layout—placement, routing, clock tree synthesis, and timing closure. This step is highly iterative and depends on the target frequency and process node.
• Mask set: The set of photomasks used to pattern each layer of the chip. The cost scales with process node and die size.
• Prototype wafers: A small production run (usually 25-100 wafers) to test the chip in real silicon. This is where most bugs are found.
• Packaging and testing: The chip must be packaged into a physical component and tested for functionality and speed. This includes the cost of test sockets, handlers, and the test program itself.

Process Node and Its Impact

The most significant cost driver is the semiconductor manufacturing process node. Smaller nodes (e.g., 7nm, 5nm) offer higher performance and lower power, but they cost dramatically more:

• 180nm-130nm: Mask sets cost $50,000-$200,000. Suitable for simple, low-performance chips.
• 90nm-65nm: Mask sets cost $200,000-$500,000. Common for mixed-signal and RF chips.
• 40nm-28nm: Mask sets cost $1-3 million. Used for mid-range digital ICs.
• 16nm-14nm: Mask sets cost $5-8 million. Requires FinFET transistors.
• 7nm-5nm: Mask sets cost $10-20 million. Only viable for very high volume or high-performance computing.

In addition, the engineering effort increases at smaller nodes because of more complex design rules, timing closure, and physical verification. A design that takes 100 engineers on 28nm might take 150 engineers on 7nm.

Estimating Engineering Costs

Engineering cost is usually estimated by headcount and duration. A typical ASIC team includes:

• Architects (2-5)
• RTL designers (5-20)
• Verification engineers (10-30)
• Physical design engineers (5-15)
• DFT (design-for-test) engineers (2-5)
• Project management and software (3-10)

Assume an average loaded cost (salary + benefits + tools) of $150,000-$250,000 per engineer per year. A small, simple ASIC (e.g., a power management IC) might take 20 engineers for 12 months, costing $3-6 million. A large SoC (e.g., a smartphone application processor) might take 300 engineers for 24 months, costing $18-36 million.

Verification is often the largest single line item. Many projects allocate 50-70% of engineering time to verification because a single bug can cause a respin costing millions. Using formal verification, UVM testbenches, and emulation can reduce risk but adds tool licensing costs (often $1-3 million per year for a full suite).

Estimating Manufacturing and NRE Costs

Manufacturing costs include mask sets, wafers, packaging, and test. Here is a practical approach:

1. Estimate die size: The die area is roughly the total gate count divided by the gate density for the process node. For example, a 10 million gate design at 28nm might be 20-30 mm².
2. Get mask set cost: Contact a foundry (TSMC, GlobalFoundries, etc.) for a quote. Use the ranges above as a starting point.
3. Estimate wafer cost: A 300mm wafer at 28nm costs around $3,000-$5,000; at 7nm it’s $10,000-$15,000. The number of dies per wafer depends on die size (use a die-per-wafer calculator).
4. Add packaging and test: Packaging can be $0.10-$5 per unit depending on pin count and package type. Test adds $0.05-$1 per unit plus the cost of the test program (engineering time).
5. Add prototype run: For the first silicon, you typically order 25-100 wafers. This gives you a few thousand chips for bring-up and validation.

For a 28nm chip with a 30 mm² die, a typical NRE breakdown might be:

• Mask set: $2 million
• Prototype wafers (50 wafers): $200,000
• Packaging and test setup: $100,000
• Total NRE: $2.3 million

Then the per-unit cost at 100,000 units would be roughly $10-15 for the wafer, $0.50 for packaging, and $0.20 for test, so about $11-16 per chip. At 1 million units, the NRE is amortized to $2.30 per chip, making the total around $13-18 per chip.

Common Cost Pitfalls and Misconceptions

A common misconception is that the mask set is the only NRE cost. In reality, engineering labor is often 2-3 times the mask cost. Another pitfall is underestimating verification time—many projects miss deadlines because verification is 70% of the work, not 30%. Also, do not forget the cost of software tool licenses (EDA tools), which can be $1-5 million for a full flow at advanced nodes.

Another pitfall is ignoring the cost of a respin. If a bug is found after fabrication, you must pay for a new mask set and new wafers—potentially doubling your NRE. This is why verification is so heavily emphasized. Also, beware of hidden costs like design for manufacturability (DFM) checks, reliability testing, and qualification for automotive or medical standards, which can add months and millions.

How to Build Your Own Estimate

To estimate your specific project, follow these steps:

1. Define the chip’s functionality and performance targets (clock speed, power, I/O).
2. Estimate the gate count or transistor count. Use known reference designs or IP blocks.
3. Choose a process node based on performance, power, and cost trade-offs.
4. Calculate die size using gate density figures from the foundry or literature.
5. Estimate engineering effort: number of engineers × months × loaded cost.
6. Get quotes for mask set, wafers, packaging, and test from foundries and OSATs.
7. Add a contingency of 20-30% for schedule slips and respins.

For a quick ballpark, use the rule of thumb: total NRE = mask set cost × 3-5 (to cover engineering). For example, a 28nm mask set at $2 million gives an NRE of $6-10 million. This matches many real-world projects.

Alternatives to Full Custom ASIC

If the cost is too high, consider these alternatives:

• FPGA: Field-programmable gate arrays are slower and use more power, but have zero NRE and can be reprogrammed.
• Structured ASIC: A middle ground where the base layers are pre-fabricated, reducing mask cost and turnaround time.
• Standard-cell ASIC with older node: Using 180nm or 130nm can cut mask costs to under $200,000, but you sacrifice performance.
• Multi-project wafer (MPW) shuttles: Share a wafer with other designs to reduce prototype costs.
• Buy an off-the-shelf SoC: Many applications can use a commercial processor or microcontroller.

Frequently Asked Questions

How much does it cost to develop an ASIC from scratch?
For a typical 28nm chip, expect $5-15 million in NRE (engineering + masks). For a 7nm chip, $30-100 million. The exact figure depends on complexity, team size, and verification effort.

Why is verification so expensive?
Verification requires writing testbenches, running millions of simulations, and using expensive EDA tools. It also takes more engineers than design. A single bug can cause a $2-10 million respin, so companies invest heavily to catch bugs before fabrication.

Can I reduce ASIC development costs?
Yes, by using an older process node, reducing die size, reusing IP blocks, and using MPW shuttles for prototyping. However, these choices may increase per-unit cost or limit performance.

What is the difference between NRE and per-unit cost?
NRE (non-recurring engineering) is the one-time cost to design and tool the chip. Per-unit cost is the cost to manufacture each chip after the design is done. Total cost = NRE + (per-unit cost × volume).

How long does ASIC development take?
Typically 12-24 months from specification to production, depending on complexity and process node. Advanced nodes (7nm and below) take longer due to more complex design rules and verification.

Is it cheaper to use an FPGA instead of an ASIC?
For low volumes (under 10,000 units), FPGAs are usually cheaper because there is no NRE. For high volumes, ASICs become cheaper per unit, but the break-even point depends on the design’s complexity and the FPGA’s unit cost.

Related

• Kingdom Hearts Story So Far: Full Game List
• Galaxy S6 Edge Factory Reset: Step-by-Step
• Best Turn-Based Strategy Games Like Zero Company for Steam Machine