How Do You Decide Between Analog Chip Manufacturers?

Question

I'm an electronics beginner and I find myself often needing to pick an IC for an analog application with no prior knowledge.

In the domain of microcontrollers, it's very clear that I should use Atmel AVR since many hobbyists use the ATMega8/128/168/32u4 etc. due to Arduino + I've got the toolchain setup for that now. I've familiarized myself with the line at this point and I don't bother looking elsewhere.

But when I'm researching something like signal conditioning, I'm always finding myself asking: so do I get an IC from Analog, Linear, TI or Microchip? Especially these three companies have a lot of overlap in the signal conditioning domain, but I'm sure you can find other groups of companies with similar overlaps in other domains.

When you're picking an IC, do you have a favorite or "go to" company that you'll start with? Or do you compare the specs from all the main companies and go with the "best" one? Do you leave it up to chance and go with the first thing you find that satisfies the requirements?

Do things like the general quality of datasheets, customer service, website UX and marketing influence your decisions?

Answer 1

There is no single rule that drives you to just one analog chip manufacturer's parts. (Not in microcontrollers, either, despite your claim that only Atmel makes sense for you. But that's a separate question.)

A really good component engineer could probably fill books about how to choose a chip and/or manufacturer. I am not a component engineer, but I can lay out a bunch of cases where I've been forced to one manufacturer or another:

• Unique chip: TI's TLE2426 is truly unique. There are numerous ways to build something similar out of generic components, but if you need that function and it has to fit in a TO-92 footprint, you're stuck with TI's chip.

• No standard pinout: Unlike with op-amps, analog buffers never really settled on a common pinout, and there are few enough of them available that it's rare for more than a few to be really suitable for a given circuit. Say you pick NatSemi's LMH6321 for some reason. Even though another chip may be an acceptable alternative, no other analog buffer on the market shares that particular pinout, so as soon as you make a PCB with the LMH6321 in mind, you cannot change to another chip without re-spinning the board.

• Subjective matters: In high-end audio circuits, a good ear can distinguish the sound of op-amps, or at least, op-amp families. If you happen to like the Burr-Brown sound, you will be looking only at their chips, while another will swear by Analog Devices chips, and you may both wish to ignore the Linear fanbois.

• Price: Maybe you need an LM324 but aren't too picky about quality. In that case, you're going to be looking at one of the second-source suppliers for this chip like Fairchild or ST, rather than the chip's creator, NatSemi.

• Durability: This is the inverse of the previous point. I'll bet anyone in the industry for more than a few years could tell you stories where the design engineer chose something widely second-sourced (an LM317 for example), did all their prototyping with a high-quality implementation of that chip design, sent it off to the manufacturing folk who substituted a far cheaper generic, and then had to deal with field failures because the replacement, though claiming equivalent performance in its datasheet, turned out less durable when abused in ways you cannot control from your clean lab. (ESD in winter in North Dakota, idiot end users unplugging connectors the manual says not to unplug while power is applied, etc.) It is often cheaper to go back to using the more expensive first-source version than re-spin the board with protection components scattered around the cheap knockoff.

• Obsolescence: You're tasked with stuffing boards you can't re-spin for a design made 20 years ago, and some of the chips aren't available from their original manufacturers any more, or at least aren't available in the packages the design engineer chose 20 years ago. You might be forced into the arms of a company like NTE or Central Semiconductor, who specialize in offering old designs that all the tier 1 companies have abandoned.

• Higher performance: A highly popular IC like the LM317 often inspires pin-compatible follow-ons with higher performance. Linear Technology specializes in this. If you find yourself needing better performance, it is often worth paying the higher price for a less common pin-compatible replacement than redesigning your circuit to give better performance with the original chip you selected.

• Availability: Once you've run the gauntlet above, you might still seem to have a few choices of sources for a given part, but one of the sources has burned you in the past on availability. That sort of thing can force you to another manufacturer all by itself.

  A more extreme variant of this situation is that you've whittled away your choices until you end up looking at just one part, but it's made by a manufacturer that doesn't always have ready stock. That might force you to abandon that choice, desirable though it may be on paper, simply to switch to a more reliable source. You might even end up reinventing that perfectly good wheel with generics just to get better control of parts availability.

Answer 2

Completely depends on the application as 'analog IC' is a massively broad term.

It comes down to price, performance, package, environmental rating, required external components, availability(location, volume, distributor), etc.

If your coming at it from the hobbyist perspective then your choice is usually driven by low unit count availability (e.g. can I buy just 1 from digikey?), easy to breadboard package, example usage from some other source, is the thing stable with minimal breadboard-able external components?, etc.

You can't just look at the numbers on a datasheet and expect it to 'just work' when tossed into a design. Some devices will work like this, some will work poorly, some won't work at all. I've seen/heard of lots of hobbyists go to digikey and buy some 50Mhz high end super duper opamp and toss it in a breadboard only to spend the next 2 weeks trying to figure out why their circuit oscillates. Another example is going out and buying what looks like a similarly specced part at first glance but the part has a different mode of operation which causes small, but significant changes in the required implementation that are often overlooked.

So one important metric is how resilient a part is to 'poor' designs. I don't mean to sound insulting, but there really is a lot more to analog design than copying an opamp topology off wikipedia. Simple, highly stable parts will allow such an approach but generally have limited performance envelopes.

I've seen production designs have this issue as well. e.g. It worked on the prototype but no one looked at the circuit stability and it was teetering on the edge such that in the first test run 5 out of 100 failed or there are intermittent failures that seems to happen completely randomly. Another example is a prototype run was done but someone didn't properly consider input/output impedance, input current, quiescent current, required decoupling, etc.

In summary the metrics used vary widely by application and needed performance. For hobbyist use its best to stick with simple, stable and resilient ICs that come in easy to hand work packages which have been used in other designs you've seen on the internet. Even so you may run into troubles just copying the design. I've seen the contact capacitance in breadboards throw all sorts of seemingly simple op amp circuits into oscillation.

Answer 3

• Second Source Available: When you use parts available from multiple manufacturers, you have a pretty good way to get ICs that are industry standards. In mass production, as Madmanguruman has said, this means less trouble when certain manufacturers tell you amazingly long lead times; for the hobbyist, it means that your chances to get the part from small-volume distributors are far better.
• Package options: Another way to find good mainstream parts is looking for package options. If an op amp is available in CERDIP-8, DIP-8, SO-8, MSOP-8 and whatnot (maybe even metal can TO-99), you can be sure it is a widely available part used successfully in plenty of different applications and likely far from being obsoleted within the next couple of years.
• Good and detailed data sheet available: Read the data sheet in detail and compare data sheets published by different manufacturers. Buy the part from the manufacturer whose data sheet you trust most. I've come across many data sheets that even have conflicting information within themselves, and those were the parts I've often had trouble with.
• Cost: Not that the difference between 20 and 50 Ct. will kill you when you only need four or five ICs for your project, but if you find similar parts at different prices, the cheaper one is usually the one closer to mainstream. You can also look for the "product status" category on the manufacturer's website: If they tell you a part is "active" that equals good and if they tell you it's "not recommended for new designs" it means that you should not buy it unless you need it to repair stuff.

Answer 4

Where possible, choose ICs that are available from multiple vendors. This means pin- and function-compatible. Most manufacturers ensure that all part numbers used on mass-produced BOMs (bills of material) are multiple-vendor parts unless specifically approved. There are situations in which unique parts are needed, but minimizing the number of parts like this is an important supply chain management function.

This puts more work upfront, qualifying multiple parts during the design cycle, but it's better to do this early rather than once something is being mass-produced and the line is down due to a parts shortage.

Answer 5

All of the other answers are fairly good, but you all forgot one important thing: simulation! Analog circuits are often somewhat tweaky, and spending a little bit of time with a simulator can often save hours or days of headaches later on.

Many of the manufacturers have some level of support for simulations, but they all have issues. In my opinion, Linear Tech's LTSpice is probably the best free tool available for simulations.