I've had some additional thoughts on creating integrated analog ICs. These thoughts came to mind following the publication of two of my blogs on debugging custom ICs and putting a DAS on your PCB to make it easier to trouble-shoot remotely; and Bill Schweber's blog on debugging tactics.
Sometimes, it may be more appropriate to use individual ICs that contain small- or medium-scale integration IP (intellectual property). Consider an example: Let's say you're designing a product that is a small stereo (2-channel) audio processing device for a consumer application. Let's assume it will accept several audio inputs, process the audio, amplify the audio, and drive two speakers.
Let's further assume that the “processing” means a multi-band EQ, balance control, and volume control. And you'd like a very simple user interface — perhaps a few push-buttons to increase and decrease the settings. That's all you need , but you wouldn't object if you could get a few more features thrown in for little or no additional cost — like reverberation, bass enhancement, and spatial separation tweaking.
After a look at various manufacturers' sites, you see that a few companies have the IP you need and the extra features that you would like thrown in. The per-unit cost is quite reasonable. You would be able to build your audio device on a tiny PC board and would just need to add connectors, speakers, and a power supply.
But wait! There's a significant NRE (non-recurring engineering) cost. That's the one-time cost that the IC company is going to charge you to design and test your circuitry on a chunk of silicon. If you were planning on building 1k or 10k units, that NRE amortized over a production run that small would make the per-unit cost exorbitant.
So, in an example such as this, you may be better off picking your op-amps and class-D amplifier ICs, laying out a larger PC board, and building your product the old-fashioned (i.e., pre-21st century) way. There is an added benefit: Trouble-shooting your device (both at the design and manufacturing stages) is far easier and less costly.
Have you encountered a problem like this before? At what manufacturing quantity did you hit the break-even point (integrated version vs. individual ICs)?