mrb's blog

ASIC Development Costs are Lower Than You Think

Keywords: bitcoin hardware

The minimal development costs for an ASIC are much lower than people think. These costs go toward:

  1. Designing the ASIC; the output is typically a GDSII file (a one-time cost: engineers salary + software licenses).
  2. Producing the mask set at the foundry (a one-time cost).
  3. Producing the wafers at the foundry (typically a fixed wafer lot cost + a cost per wafer).
  4. Slicing the wafer, and packaging the chips (fixed cost per wafer or per chip).

The cost of #1, designing the ASIC, is the most variable. It can be literally a few thousand dollars for a simple ASIC (eg. a Bitcoin mining ASIC with simple repetitive SHA-256 units), up to billions of USD (eg. the Cell processor development which cost 2 billion USD). In the case of a simple ASIC, the cost of #2, the mask set, is the biggest. #1 and #2 are usually refered to as NRE (non-recurring engineering) cost.

So what is the total cost for designing and producing a small initial batch of ASIC (sum of costs #1 through #4)? Let's hear from real-world sources who did produce Bitcoin mining ASICs:

  • The Bitfountain company said: in China, the cost is ~150k USD for 130nm, and ~500k USD for 65nm. They ended up developing a 130nm ASIC.
  • The BitSyncom company, also in China, posted enough information to reveal that their cost at the TSMC foundry were 200-300k USD for their 110nm ASIC. This matches the amount they raised via crowd funding (300 units * $1200 each = $360k).

So, there you have it, from two horses' mouths: ~150k USD for 130nm, 200-300k USD for 110nm, and ~500k USD for 65nm, as of 2013.

Comments

Joachim wrote: Are these numbers just the NRE cost, or the total cost for X number of produced chips? 20 Apr 2013 11:34 UTC

mrb wrote: These numbers are the total to produce an initial small batch of chips (so costs #1 through #4). Foundries usually require the minimum batch to be 6 wafers. 21 Apr 2013 08:17 UTC

dan wrote: Hi, can you find me the engineers and software if I have $150K to produced the 130nm Asic chip ? Thanks 26 Jun 2013 06:39 UTC

mrb wrote: dan: China is your market. You won't be able to do it for $150k in the USA or Europe. 15 Nov 2013 20:31 UTC

brian piercy wrote: Unfortunately the examples don't provide a measure of the ASIC's complexity. Pre-tapeout expenses (designers' headcount) will be a function of how many testable features are included on the chip. In my experience I've NEVER seen an ASIC project team complete a design for less than $500K USD, regardless of location. 19 May 2014 02:44 UTC

mrb wrote: brian: you are right. As I explained in the post, there are 2 extremes: a dumb ASIC like a Bitcoin mining core (which is in a way self-testing[1]), or a full-fledged new processor like the Cell. The vast majority of ASIC projects are a lot more complex than a Bitcoin mining core, so you will end up spending more than the numbers in this post. My point was to document what the basics costs are for an ultra-simple ASIC.

[1] If it finds SHA256 solutions at the expectetd rate, then it is 100% working.
23 May 2014 01:33 UTC

orphanblok wrote: Would love to read an updated version of this now that it's four years later and we're at 14nm.

Highly specialized information like this (and comment section) will be very important to continue improving decentralization.

Maybe one day we can have a Open-Compute Bitcoin ASIC project.
18 Mar 2017 18:59 UTC