I've always felt there was little need for considering analog application-specific ICs (ASICs), with all the uncertainties and risk factors which I assumed they involved. Why go to all the hassle, when you can go to the web sites of dozens of reputable analog IC vendors, find a “near-perfect” combination of ICs you need, and put together a circuit that mixes processes and performance while minimizing the tradeoffs?
For example, you can combine a high-end, super-low-bias instrumentation amplifier with an appropriate variable gain amplifier, ADC, some filter-function ICs, specialized line drivers, and any digital I/O you need (UART, SPI, I2 C) into a circuit that should do your job with minimal headache. Yes, there are layout and power-supply issues (noise on the rails, overall dissipation) but those should be manageable if you follow good design practices. Plus, there are vendor reference designs and application notes to help, as well.
So, again, I wonder: why even consider the analog ASIC? It has too much risk, too little flexibility, and you are not leveraging the access to the tens of thousands of excellent analog parts available as standard, quick-delivery items.
But sometimes you have to at least listen to, and consider, the other side of a story, of course. Perhaps there are some assumptions you are consciously or subconsciously making perhaps were never true, or are no longer true due technology advances. Perhaps there are factors you hadn't even considered that should be part of the decision matrix.
This image is courtesy of Cactus Semiconductor. See James McDonald’s blog The Enabling Chip Technologies Behind Miniature Implantable Medical Devices
That's why I read a recent article in Medical Design Briefs with interest. The piece “Optimizing Electronics for Medical Applications” lays out eight reasons why analog ASICs make sense. Yes, the author is a vendor of these ASICS and so is biased, but so what? His points may still make sense, and good engineers can judge the points and adjust for that bias. In brief, he makes these points:
- NRE (non-recurring engineering) costs may be in the tens of thousands, but they can also go toward zero if the ASIC volume is high enough and the ASIC vendor wants the business;
- While development times may indeed be months for a full-custom analog ASIC, they can be as short as a few weeks if you go with a tile array;
- Counterfeiting/copying issues are greatly minimized, since there is no public datasheet, no PC board that can be reverse engineered, and the IC is sold only to you;
- Reducing the number of ICs to one or a few ASICs not only reduces required PC board real estate, but also saves space due to reduced need for passives such as bypass capacitors;
- It can be very cost effective in terms of the BOM and assembly costs;
- Reliability increases, also due to fewer ICs and interconnections;
- Special shapes of the IC die are practical, such as for a unique aspect ratio so the end-product fits inside a catheter; this may make some products feasible that are otherwise not possible;
- Obsolescence is far less of an issue, since you are not at the mercy of “final buy” notice from one of your IC vendors; this is especially important in markets that routinely expect sell the same product for five, ten, or more years, especially if there is regulatory approval involved.
I still think that analog ASICs are only a good idea in limited cases. Perhaps that's my own semi-rational bias, because I like to come up with a circuit design, order the parts, have them delivered within a few days, and start the breadboarding and prototyping phase. Plus, using discrete ICs gives me wide flexibility when marketing redefines the product objectives during the development phase, or you find you need more resolution from that op amp or ADC, or you realize you need to reevaluate some performance goal due to your own misjudgment.
Still, the points that the article raised are very interesting and valid, especially for some markets and applications; it would be foolish and short-sighted to not at least give them some consideration.
Have you ever considered using an analog ASIC instead of discrete analog ICs? Did you go that route? Why or why not?