The other day, a reader was surprised when I claimed that anyone could design and manufacture a custom analog IC for about $3,000. I suppose it does seem hard to believe, but it's true. I've done it a few times and had great success every time.
Granted, I've been designing analog ICs since about 1984, but then university students do this on a regular basis. The concept relies on a fairly old crowdfunding model. The nuts and bolts of the process go something like this.
First, you need to acquire some schematic, simulation, and layout tools. These are available for free — LTSPICE, MAGIC, Electric, etc. A great resource for locating free IC design tools is Fedora Labs, which even has great PCB tools.
Then you need to pick a process from one of the multiple fabrication houses that support what is called a multi project wafer (MPW) service. You will need to set up an account with the chosen fabrication house/broker, and that requires signing a nondisclosure agreement. Then you can download its process design kits, which include all the simulation models, some layout examples, bonding pads, and the design rules for the process.
After you have designed, laid out, and verified your circuit, you hand it over to the manufacturer to be merged with other designs from around the world to share the manufacturing costs for a single run. After you hand over the data, it takes up to four months to get your parts in the mail.
There is only one time in this process where money is required. That is when the chips are manufactured and packaged.
There are several MPW manufacturers and processes to choose from. Some popular foundries (or brokers) are X-FAB, CMP, MOSIS, and LFoundry. You can design with a multitude of technologies, but it is best to stay away from the most advanced process nodes (like 90nm) and use a fairly old technology (like 0.35um). This is for two reasons:
- The design with an advanced process is substantially more complex and requires understanding nuances in the device types that are not included in the models (for example, MOSFET gate leakage current is not zero at 90nm).
- The price for prototyping goes way up, primarily because the manufacturer must pay highway-robbery prices for the masks used in the process.
A recent check with CMP shows that a 3mm2 die will cost about €1,950 ($2,550) for 25 parts. All MPW houses have fairly similar prices and are located around the world. One can fit more than 100,000 transistors in a 3mm2 die. For analog, that is more than enough.
Though €1,950 is not cheap compared to a PCB run, one can always partner with a few friends and split the price. Universities also get a price break; you might check with your local university and see if you can ride along. If you do share the silicon real estate, just make sure you don't share the pins, power and grounds, or signal lines. Think hotel — separate rooms but a shared lobby. Just imagine if your friend put in a power ground short and killed everyone's project. You may lose your golfing partner.
All MPW manufacturers will manage packaging for the run. This usually means about €1,000 for 10 packaged parts. Generally, all of the money is due up front.
Free design tools can be found, but you might want to consider spending $100 on at least the layout/schematic tools (like Juspertor) to simplify verifications between your schematic and your layout. These generally involve layout versus schematic, design rule, and electrical rules checks. A coupled tool will allow you to cross probe the layout errors to the schematic, extract layout parasitics, and add them to the simulation netlist. You won't need to go back and forth constantly between multiple tools. There may be free tools available that are coupled, but none come to mind.
I've glossed over the finer details, but this is the general flow. One note of caution: IC design is not the same as PCB design in one key way. The former is for perfectionists, and the latter is for people in a hurry. It is easy to swap out caps on a PCB if you chose the wrong value. On an IC, it will take another €2,950 and four months.
I liken troubleshooting an IC to how a doctor troubleshoots a medical problem. When you come in with chest pains, he doesn't immediately slice you open and start swapping organs. Lots of tests are conducted from the outside, and then the doctor makes a highly educated guess on what to do on the inside. Sometimes you're right, and sometimes you're wrong, in which case you keep looping in the process until you get fired or you get frustrated and quit. IC design is not for the faint hearted.
If there is enough interest in the topic, I can blog a bit more about the details. Let me know.