Part numbering etc.

I find programs to be clunky for something as simple as even a complex spreadsheet because the spreadsheet is both program and development system in one. Otherwise you get all the complexity of programming with source, an editor, an IDE, someplace to keep the source files distinct from the data files, some form of output from the program... etc. All this has to be maintained. With a spread sheet you have the tool and the data and possibly the output if different from the spreadsheet.

I track all my production, invoicing, accounts receivable, accounts payable, etc. in one spreadsheet. I find I can see everything I need if it is constructed well. Conditional formatting gives me warnings or highlights and can hide unimportant details. I love spreadsheets.

I guess with such low volumes the optimal methods are very different. I've thought of designing a couple of low priced units, but I would want to handle the entire manufacturing, sales and shipping via some far east outfit that gets the great prices of all of that. I've never found a provider for that sort of service. Sort of an Alibaba or eBay by proxy.

Too much typing but the manufacturer's part number would also be helpful. Often there are different parts that are close to each other, or after one company gobbles up another. Then again, some companies don't care what 100ohm(...) is used. All manufacturers are the same.

Too much typing, too much space on the BOM, and makes a lousy database key.

You do have to be careful of parts aging beyond their "best if used by" dates. Unless you have ovens, this goes double and triple for semiconductors.

I've almost given up trying to keep the numbers down. At my PPoE they were really concerned about the number of parts per side. We went to extremes to get like parts on the same side of the board so it would fit into one pass of the pick-n-place tool.

No one worries about it where I am now. The layout engineers couldn't care less about the number of components on a side. Manufacturing doesn't care, either, though the CM I use for my stuff does charge a premium because of the number of parts (normally over a hundred part numbers, many mixed on both top and bottom). Different markets, different business models.

We keep parts in ziplock bags with dessicants. Big chips go into a drying cabinet for a few days before place+solder.

We try hard to not put parts on both sides. And to use one feeder setup.

The feeder count and such are yet another design constraint. It's amazing how complex electronic design can be.

Not good enough. Solder-proof parts can ruin your day. Resistors, and most capacitors, are too cheap to take chances with.

This should be done by MSL numbers. Put it in the part number database. Still, parts age and it has little or nothing to do with moisture.

I don't think I've ever done a single-sided board, other than a breakout board (jumper between mother/daughter for debug) or other simple test board. Then, they've always been part of another assembly.

Exactly. A hundred+ parts isn't unusual. It's just not possible to knock it down significantly.

We have one VME module with 1200 parts. Lately I've been using quad resistor packs in all sorts of ways, which really keeps the parts count and BOM count down. And makes it harder to fix math mistakes.

Most of my boards are in that range, but in a lot smaller space than a VME module. There are some really large components (the one I'm designing now has a pair of 100uH 15A torroids, and four 3300uF, 35V AL caps) on the board, too, making it pretty dense. If I had a third side, it would be filled. ;-) I do use resistor packs but for what they're designed for and not as many as some of my colleagues. Except for regular stuff like QFPs and memories, R-packs tend to screw up routing for little space savings.

These are inspection failures. SMT resistors and caps on end. Backward diodes. Missing components. Improper components. And yes, solder bridges. We had very extensive testing. This was a large Laser manufacturer. Some of it Medical. So nothing went out the door that was not functioning 100%.

I use them a fair amount in things like bias networks and diff amps. The diode sampler board we're doing right now has about 6 of them.

Cheers

Phil Hobbs

Well, optimize the pins for best routing.

Here's an example.

You can do all sorts of divider ratios, supply voltages, and opamp gains with quad Rpacks. Just don't change your mind about things later.

Sure but that's difficult on BGAs, where the functions are pinned out, seemingly, at random. I use a lot more on QFPs/QFNs.

I find they're not accurate enough for most things, outside of terminators and the like. The ones that are, are expensive and huge. I'd love to find a .1% 8-pin 0408 4R R-Pack.

Yikes. 58 characters!

Our BOM calls out a 7-digit company stock number. The stock record includes the descriptive stuff and lists up to five manufacturers+part numbers that can be purchased to go into the bin. If a part turns out to have a problem, we can remove it from that authorized buy list. I'd hate to create part records based on manufacturer's (often weird) part numbers.

I've seen part numbers assigned to components and assemblies at different levels. Normally the part number corresponds to a set of documentation that defines how to make that item. So each level of assembly has a description of how to make it from either components or other assemblies.

I've never seen a company use a part number to distinguish a tested unit from a built but untested unit. I suppose if you were going to build units but not test them until you needed to deliver them having two part numbers could be useful. Because of the tax many states have on inventory, the difference in taxable value of the tested and untested units might be significant.

We're required to be able to trace all parts (the Note-7 debacle couldn't happen - at least the first recall), so each manufacturer has a separate part number. No, using the manufacturers part number as a stock number wouldn't be so great but in our system there is a 1:1 correspondence. By design, there aren't any one-to-many mappings but accidents do happen (noted with a '*' P/N suffix, IIRC).