For the past couple of weeks, I have been participating in a LinkedIn discussion group topic regarding how one emulates a low temperature coefficient (TC) resistor on an IC when the IC process doesn't have a low TC resistor available. This discussion is a great example of the differences in how one designs an analog circuit on an IC versus on a PCB.
The discussion started with a detailed post by Jean-Francois Debroux. He showed how one could take two resistor types with large but opposite polarity TCs and combine them to produce a lower effective TC. The discussion split into two camps. On one side is the group that found the concept intriguing with promise. On another side are those who are skeptical regarding either the practicality or the necessity of such a solution.
While I sided with the later of those two groups on both points, the more important issue is why I took those positions.
When an IC is manufactured, there is very little that is tightly controlled with regards to absolute precision. Whereas the photolithographic processing can yield structures that match to better than 1 part in 1000, the addition of impurities into other impurities to build, for example, a resistor, is not well controlled. Absolute resistance control is limited to about 250 parts in 1000; a factor of two hundred fifty (250) times worse than relative control.
The short message here is to always design integrated circuits whose operation is based on relative principles, not absolute precision.
This fundamental design principle is taught early on to students learning how to design analog ICs. And it applies to virtually all integrated components on an IC, not just resistors.
But as Jean-Francois pointed out, in his experience, sometimes one has no choice. He cited for example the sense resistor inside of a current sensor. Another member of the group cited a case of building a precision low power oscillator. These examples lead me to the second point of disagreement: necessity.
Something I learned many years ago from several of the masters in analog IC design is that one should always work with the inherent weaknesses on an IC rather than fight them. A good example of this IC design principle is practiced regularly in the form of a bandgap voltage reference.
The principle behind a bandgap reference is that if two matched bipolar transistors are driven such that their collector currents are equal, their base-emitter voltage difference equals
where k=Boltzmann's constant, q= charge of an electron, T= absolute temperature in degrees Kelvin, and n is the ratio of emitter areas of the matched bipolar transistors.
This is a very powerful result. Equation 1 says effectively that every IC process on the planet that has bipolar transistors can produce exactly the same voltage if the layout of the matched transistors embodies an emitter area ratio of n. Typically, one finds n equal to 8 as a popular choice. This then results in ΔVbe = 53.4mV at 25°C.
Another important aspect of ΔVbe is that it is proportional to absolute temperature (PTAT) with a temperature coefficient of about +3300ppm. As it turns out, the aluminum metal interconnect used on an IC also universally has a TC of about +3200 ppm. So now one can build a current sense resistor out of aluminum interconnect and compare the current sensed voltage against the ΔVbe voltage with a near zero TC drift of an absolute over current detection threshold.
The current detector circuit description above is an example of using the fundamental aspects universally available on all IC processes to create valuable repeatable solutions without depending on absolute controls. While this is only one example, it has been my experience that if one looks hard enough or allows some freedom in how a solution is implemented, it shouldn't be necessary to design on an IC with absolute principles. Equivalent or better results are available with more reliable IC appropriate methods.
I have no doubt that from time to time, weird situations present themselves where engineers are pushed to do things that entail risks that they would otherwise prefer not to take. But from my perspective, the more important lesson here is that those who drive decisions regarding what is to be integrated need to also recognize that there are two completely different analog design methodologies at play.
To force one platform methodology onto an incompatible platform is opening up a Pandora's box of potential problems. Yes, you might succeed in forcing the issue, but you very likely will then also suffer a large consequence when product yields fall off one week, or you discover that the test program now has to be run at multiple temperatures to validate things not guaranteed by the factory.
Designing an IC using relative design principles is the universal unspoken agreement between those who design on an IC and those who control the manufacturing of the wafers.