Re: Opinions re: MCU vendors [long]



rickman wrote:
On Jan 17, 5:19 pm, D Yuniskis <not.going.to...@xxxxxxxx> wrote:
Hi,

I'm looking at a couple of different parts from Atmel
and TI for an upcoming design. I'd appreciate comments
(not *rants*) folks might have about both firms in terms
of:
- quality of their products

That is a pretty poorly defined metric. It reminds me of the Volvo

It is poorly defined only in the sense of how it is *measured*.
But, you *know* a quality product when you see it. You also know
*crap* when you see that!

commercial many years ago when the US makers were still building
crap. It was a huge presentation room with hundreds in the audience.
The speaker talked about quality and then started chanting, "Quality,
Quality,..." and the audience started chanting with him. The point
was that the US makers were just chanting quality without doing
anything measurable. But then quality is not a measurable thing. Can

Sure it is! Perhaps not in some "internationally acceptable
interchangeable unit of measurement".

E.g., I spent a few summers working for a (huge) hand tool manufacturer.
Part of my job was quality related. Judging the quality of their
current products and how they compare to other manufacturers of
hand tools.

Tell me, how do you *measure* the quality of a hammer? A screwdriver?
A tape rule? A saw??

First, you verify it meets all of your mechanical specifications.
E.g., what is the draft angle of the two faces of the (slotted)
screwdriver's tip? If it is a "cabinet tip" screwdriver, are the
*sides* of the tip parallel? Are the flutes in a Phillips
screwdriver free of buildup from the plating process? Are the
markings on a tape rule legible and accurate?

Second, you verify that it meets the other design specifications.
E.g., what is the (Rockwell) hardness of the shank? How thick
is the plating?

Third, you verify that it meets the appropriate "bogo-units".
E.g., if the shank is held fast, how much torque is required
to "strip" the handle off of the shaft? If the screwdriver
tip is pushed into a precision die with a force of X, how much
torque is required to *sheer* the tip off? How many times
can you "bang" a hammer before the handle fails?

Of course, the third group are, by far, the most fun to devise
and implement. They are somewhat arbitrary. Bogus. Yet,
they each pertain to QUALITY FACTORS that the USER WILL PERCEIVE.

Ever been pissed off because you *tore* the tip off a Phillips
screwdriver? (the average man can do this *easily* for a #0
screwdriver, "with some effort" for a #1, and "only if really
determined to do so oand with the assistance of leverage
enhanceers" for a #2. #3 requires hydraulic assistance. :> )

Notice how *long* the finish on a tape rule lasts? Despite
the fact that it is *continuously* experiencing friction?
Ever wonder why the plastic case on that 25' tape rule doesn't
shatter when the tape retracts from its full extension?

How do we measure quality in *software*? Count the number of bugs?
Are all bugs of equal "value" (within an application -- lets ignore
the complication of dealing *between* application domains)? I
was drawing a schematic last night. The "part editor" has a bug
in it that causes placement of pins not to track the cursor's
position. Annoying. OTOH, abandoning a newly created schematic
sheet causes the application to *crash*. Potentially very
expensive!

And, for the most part, we don't even have a deterministic way
of *testing* software -- at least I can set up a machine to
bang a hammer continuously until it breaks! (whatever a "bang"
unit happens to be)

you tell me what you mean by quality? Can you define it in some terms
that someone else could respond in a way that would have meaning to
you?

Sure, obvious things:
- parts failing to meet their published specification (you would
agree that this would be "crap"?)
- parts that *technically* meet their published specification
but not "in good faith". Especially nowadays with datasheets full
of "typ" numbers. ("Yes, typically Icc is 10mA. The fact that
*all* of the devices you have purchased from us draw 100mA is
still within the (unspecified) maximum that we publish for that
part.")
But, also, remember that their "product" isn't limited to bits
of plastic made in the far east:
- documentation that is grossly and obviously incorrect
- errata that are not kept current and/or are "hidden" for
fear that someone would "think ill" of their product

I was reviewing an SiLabs part last week. Application data in
the datasheet (sample schematic) had to be one of the worst
drawn documents I'd ever seen! As if someone had pasted
little *squares* (don't folks know that there are symbols for
things like optoisolators? transformers??) on an Etch-a-Sketch
and asked a two year old to "connect the dots". <frown>

This document speaks for your (their) component. It is
reproduced in perhaps 6 other app notes verbatum. So, each
time you see it, you remember, "Oh, crap. Not this again..."

Documentation for a reference design using that same component
had obvious flaws in the schematic. "What did you guys use
to layout the *board*? Why publish anything other than the
*actual* schematic that drove your PCB layout tool??"

Sure, these are nits. But, everyone who reads these documents
runs an increased risk of making a mistake *or* spending extra
time to sort out what is *really* intended in the documents.

Presumably, they would want customers to have *success* with
their products. This suggests they would want to do everything
reasonably possible to facilitate that success (proofreading
a document doesn't seem to be a *huge* undertaking!). So,
if something like this that is in *their" "best interest" is
treated so casually, what does that suggest about the "quality"
of their silicon? their support? etc.

- accuracy of their documentation

My experience is that both Atmel and TI have quality documentation.
TI seems to be less willing to modify their documentation if you have
questions it does not answer. Once I got Atmel to add some info to
their SAM7 data sheets to define crystal requirements. But then they
provided a table with data points at specific frequencies that didn't
say what to do in between. I pointed out the issue and they wouldn't
even comment verbally much less clean up the issue. So I still think
they are about equal.

It seems that manufacturers are moving more into just becoming
foundries -- or IP houses (ARM). Trying to get a mix between them
seems to be difficult.

- ability to meet commitments

I like this one. What sort of commitments are you talking about?

Wanna buy a Z380? :>

Production? Updates? Support? If you are a small player, don't
expect too many commitments in the first place, much less holding to
them. All IC makers have a handful of major customers for whom the
sun rises and sets. Everyone else in secondary or even tertiary.

Of course! I've bought components "by the pound". And,
size shouldn't affect their *ability* to meet their commitments.
When I'm buying 1M of a particular device / year, I have a different
level of expectation (tell me I have to wait 3 years and I'll smile
and turn a new crank on thedesign so your name isn't on any of
the components). When I'm buying 1K of that same device / year,
I *don't* expect the same *timeliness* of a commitment -- *but*,
if you tell me 18 weeks, then I expect it to *be* 18 weeks and
not 24 weeks or 36 weeks or "gee, your request must have got
lost in the cracks"!

I.e., I expect a vendor to cater to a (known) bigger client
more promptly (allocation, etc.) than a small fish. But, I
expect him to meet those commitments (that *he* is defining!)
equally.

E.g., I expect a banker to kiss the ass of someone with $1M
on deposit. Walk in the door and perhaps he even gets up and
holds your chair for you, etc. OTOH, someone with $100 in a
savings account maybe can HELP THEMSELVES to a complimentary
lollipop from the big candybowl at the teller's window.

BUT I EXPECT EACH OF THEIR TRANSACTIONS TO BE PROCESSED WHEN
PROMISED (maybe an instant wire transfer "on the house" for
the big shot and "10 days" for the small fry)

- responsiveness during the design process

Again, I have not found a significant difference between the two
companies. I have found the occasional FAE who is exemplary. I've

Agreed.

had a TI FAE who would walk across burning coals (he's no longer with

I think these folks (i.e., the ones that are "good") tend to
either have a ood work ethic or are engineers that crossed over
into "support". I'm sure it must be a double-edged sword
for them: on the one hand, you are free of the pressures of
coming up with something "never done before" under a deadline;
on the other, you never experience the depth of learning that
requires; on the *other* other hand, you get exposed to a wider
range of product ideas (which you can always mull over at your
leisure without *having* to do so). For the right sort of person,
it could be a dream job. For others, a nightmare.

TI unfortunately). My current Lattice FAE is pretty durn good
although I think he would want to good pair of boots before the walk
on coals. I want to say all the support I've gotten from Atmel was
through distributor FAEs and was other than great. Disti FAEs are
typically trained in many product lines from many makers and just
don't have the time to dig into them all in detail. But they can be

Agreed. I've had friends who have outright told me that their
goal is to "shmooze". Informal get-togethers (golf, ball games,
etc.) with clients to build personal friendships -- the product
isn't important. :<

Makes you wonder what happens to those accounts when FAE Joe
moves to another disti! :>

helpful in getting the ear of the factory support people. So you need
to gauge your local FAEs yourself.

etc.

I often have to approach vendors as a "one man shop" during
product development with production buys done by clients
*after* the design is finished. So, I usually can't carry
the clout that their buying power will (NDA's).

I use contract manufacturing and get the same result. If you aren't
buying your own parts, it is much harder for the support people to get
credit for your orders, assuming you have any. In your situation, you

Of course! I'll buy 100 pieces and they'll never hear from me again.
They'll never (by design!) connect my efforts with larger purchases
later on by Company X.

will be invisible or nearly so. It helps if you customer's
procurement people are willing to cooperate. But you need to get them
in the loop with your local salesperson. Of course this is a PITA,
but it helps to get credit to the right people and I think they really
appreciate it.

In my case, that is what they *don't* want. Often, I design a product
because a client doesn't want anyone to know they are looking into
a particular market or product offering. As much as people like to
*think* secrets are kept, you'd be amazed at how much "proprietary"
information leaks out of sales reps, FAE's, etc. I have a client
who loves to describe how he frequents sandwich shops (etc.)
located near his competitors during the lunch hour and casually
"reads his newspaper" -- all the while listening in to the
banter going on around him (i.e., engineers and sales people
from those firms chatting about work).

Look how much you can discern on USENET about the types of
products people are working on. Even if you *don't* know
which firm employs them, etc.

And, I tend to use devices in unusual ways :> So, I tend
to end up digging around in dark corners of implementations
uncovering things that they didn't expect (or *know*) would
ever occur. This is particularly important nowadays when
everything is spec'ed with "typical" numbers and damn few
"worst case" numbers ("Yikes! The power supply went into
current limit! I guess the *real* power consumption is
considerably more than the 'typ' figures you guys published...")

I.e., I don't want to waste time with firms that are either
slow to respond to *highly* technical inquiries (in the past,
I've had face-to-faces with the actual designers of many
components that I've "stetched" in unusual ways) or simply
brush them off thinking it's not worth their time/effort
(Motogorilla has lost *many* designs over the years on account
of this).

Yep, if they can't see the dollars, they don't respond very well.
Isn't that true of us all?

I like to think it isn't true of me! I try to do my best
regardless of the size of the contract or the vendor. But,
that's because I'm doing this "for me" and not just "for pay".
<shrug> YMMV.

All in all, I don't think you will see a big difference between Atmel
and TI. I think Atmel will be slightly better in terms of support.
But it also depends on the products. If you are looking at the CM3
CPUs, TI doesn't make them. They've bought Luminary Micro and I think
much of LM still works the same as it did. So you would need to check
out LM as a distinct entity from TI.

Thanks for your comments!
--don
.