7
\$\begingroup\$

I'm new to IC design and just wondering if it possible to order IC production if you have all necessary design files (like RTL design, chip layout from Cadence for example, etc.)

If it is then what are these companies and what is the approximate cost (for a relatively small IC like for example an ethernet PHY?) \$ \ \ \ \$

\$\endgroup\$
17
  • 17
    \$\begingroup\$ There are possibilities but these should be your absolute last choice compared to using something off the shelf. To be realistic, if you are asking about it here, you are nowhere near ready to consider such an option. And this questions fails the Stack Exchange requirements for specificty. \$\endgroup\$ Jun 13, 2020 at 16:27
  • 8
    \$\begingroup\$ What does your PHY do that the thousands of PHY's on the market don't already do? \$\endgroup\$
    – Ron Beyer
    Jun 13, 2020 at 16:37
  • 9
    \$\begingroup\$ The approximate cost is in the hundreds of thousands to tens of millions range. Getting a custom IC should never be your first option. \$\endgroup\$
    – Hearth
    Jun 13, 2020 at 16:44
  • 10
    \$\begingroup\$ There are perfectly valid situations where going ASIC makes sense, even as first option. If you find yourself asking this question on stackexchange, you are not in that situation. \$\endgroup\$
    – Joren Vaes
    Jun 13, 2020 at 16:58
  • 6
    \$\begingroup\$ @Chris Stratton : Even if he can't do it, that doesn't take away from the informativeness of an answer; not everything one reads about can one put into practice. And it implies that somehow, that knowledge would come about by "magic" simply by learning more about electronics. Eventually, someone has to communicate it to you. There is no logical reason that can't be here, to get the ball going, even if it never "gets" there, or only gets there many years from now. \$\endgroup\$ Jun 14, 2020 at 13:47

4 Answers 4

19
\$\begingroup\$

You'll want to check out multi-project wafer services, the best-known of which is MOSIS.

\$\endgroup\$
14
\$\begingroup\$

It is possible, but it is neither quick nor cheap.

There are several different types of fab you could consider talking to. Near me, I have two options, one is a large university which has their own microelectronics centre. They are a little cheaper and keen to try things out, but you'd have to justify why your design is interesting from an academic point of view. I would probably pick them if I want to make a handful of MEMS devices for R&D. The other is a moderately sized company which operates three small fabs and makes a point of begin able to do small runs and prototypes. They might be a better choice for a small run of chips for products to sell. You wouldn't want to start with one of the really big fabs, not unless you're willing to commit to spending a lot more.

Once you have a fab, you have to start the process of adapting your design and their processes to work together. You buy wafers, get masks made etc, test the results, and repeat as necessary. Doing a small batch of wafers, with only a few processes (so probably simpler than your PHY) costs a few thousand pounds. Developing it to the point of reliably producing 8 inch wafers full of mostly-working dies might cost £100k or so. Working up to a fully packaged IC, with all the necessary QA processes etc. might need an investment of a couple of million. These are very rough numbers, treat them as order-of-magnitude guides nothing more.

\$\endgroup\$
9
\$\begingroup\$

The short answer to ‘how’: contact fab vendors and describe what your requirements are to figure out what services you will need. They will walk you through the process and identify your costs for budgeting.

Who are these companies? The best-known wafer fabs are TSMC and Globalfoundaries. There are others that specialize in smaller wafers/older processes, like SMIC in China.

Fab companies provide various levels of service depending on your needs and how much money you have to spend. In your case you might need to work with a third-party provider who can do all the back-end work for you vs. working directly with a fab. Fabs can identify appropriate partners who do that service.

What’s it going to cost? The main up-front cost is the mask set. Three factors influence this: die size, number of layers, and process node. For a small, mixed-signal chip like an Ethernet PHY, it could use an older process like 180 or 120nm. Ballpark for such a mask set would be about $75-150k.

The next fixed expense is test fixtures and test development. This can run to multiple tens of thousands. Add to that fixtures for accelerated-life and ESD/latchup testing.

You must also do a package. Even if you choose a standard type, it needs to be laid out with an appropriate bonding scheme for your die.

There’s other expenses I’m certainly missing here but you get the point: expect up-front costs in the multiple hundreds of thousands even before you get your first packaged part.

Different companies can negotiate pricing. In one model, they can bury some of the fixed expense by increasing the cost of the delivered IC, but they will pressure you to take agreed-upon quantities.

Finally, if your design is of good quality with unique differentiation, you could possibly sell or license it to an existing company in that market space and defray some of your costs. But for it to have value, it needs to be proven in silicon. A lower-cost way to do that is to put your die on a ‘shuttle’ - a multi-project wafer - to get some sample chips for testing.

\$\endgroup\$
1
  • \$\begingroup\$ Thank you for the answer. Would multi-project wafer be much cheaper option? \$\endgroup\$
    – Andy
    Jun 14, 2020 at 7:12
2
\$\begingroup\$

You might want to check out FPGA for smaller series. Some of them contain quite a few logic blocks, and perhaps even som analog circuits.

Wikipedia has some descriptions: https://en.wikipedia.org/wiki/Field-programmable_gate_array

\$\endgroup\$

Not the answer you're looking for? Browse other questions tagged or ask your own question.