Pro EE's Which large IC manufacturers make you cringe? No Docs/No support/etc

[jakeisprobably]

 Which large IC manufacturers make you cringe? Why? Who's bad at a hobbyist level versus pro level?
   I want to know who to avoid before I learn the hard way.

EDITED:
 So after a couple of days this thread has had some interesting responses. In the interest of other beginners like myself the following is a condensed summary of the first 32 posts from this thread.

 The listing is in roughly the same order as first mentioned.
Listed: Manufacturer - Number of Opinions - Summary Explanation
Con List
Maxim- 9- Availability/Lead Times/Expensive
Marvell- 2- NDA's/Small Quantity Sourcing
Broadcom- 2- NDA's/Small Quantity Sourcing/MOQ
STmicro- 2- Datasheets/App notes
HID (RFID)- 1- Price/Sample Policy
Renesas- 2- NDA's/MOQ
nVidia- 1- Support
Microchip- 5- Quality/Price/Part Numbering/Errata
Fairchild- 2- Datasheet Incomplete or Unclear
AMD- 1- "Embedded stuff became NRND before release"
Intel- 1- 'Poor Strategy'
Qualcomm- 1- NDA/MOQ
Infineon- 1- "Microcontrollers suck big-time"
Cypress- 1- "They are weird"
Linear Technology- 1- High market price
Texas Instruments- 3- Spec Noise/Forced IDE Use/Info-management/Closed Personal e-system Support
Analog Devices- 2- Support
ON Semiconductor- 1- 'Image Sensor NDA'
Globcomm- 1- NDA
Aptina- 1- NDA
Freescale- 1- Product Availability/Lead Time
Xilinx- 1- 'Harder than Altera to use'

I didn't ask for the "Pros" side but I will just list the brand and number of opinions.
NXP- 6
STmicro- 2
Maxim- 1
AMD- 1
Texas Instruments- 4
Analog Devices- 1
Linear Technology- 3
Fairchild- 1
ON Semiconductor- 1
Atmel- 1
FTDI- 1
Freescale- 1
IRF- 1
IXYS- 1
ADI- 1

[Dago]

I tend to avoid Maxim due to (often) poor availability and long lead times for the products.

[dmills]

Seconded, some really interesting parts but you dare not design any of their unique stuff in.

Marvell/Broadcom pain in the tail for NDAs and small quantity sourcing.

Regards, Dan.

[tron9000]

thirded on Maxim, They make some pretty Schmick stuff, but you go to the supplier and find they are too expensive or are obsolete.

[JPortici]

+1 on maxim

i'll add that i would love to kick whoever writes the datasheets/appnotes (or who decided the policy on how to write those) for st parts square in the nuts

[Rerouter]

HID, yes the RFID ones,

They wont even let you buy a sample unless you sign on to buy X in the next 2 years... which is the whole point of buying a sample... 

[ebclr]

Renesas is another one to avoid

[Kjelt]

Renesas is another one to avoid

Perhaps it is usefull to explain why ?

[XFDDesign]

Jumping in on the Maxim bandwagon.

They make some incredible parts.

And do their best to not sell them to you.

At Intel, there were managers who would go cube to cube looking for maxim databooks and then disposing of them, the supply problem was that bad.

[JPortici]

At Intel, there were managers who would go cube to cube looking for maxim databooks and then disposing of them, the supply problem was that bad.

Maxim stories..
I want more! i want more!

[f5r5e5d]

is Maxim still that bad? - in the late 90's I designed in a sw C filter part, built a 1st run, it sold OK for our small niche products

but when we wanted to build a 2nd batch, a year or so later, the parts were unavailable in distribution, with official 18-24 mo lead time

[Ice-Tea]

At some point, I tried to get nVidia support on a project. Never got any. AMD was a bit better, albeit far from a breeze.

[langwadt]

is Maxim still that bad? - in the late 90's I designed in a sw C filter part, built a 1st run, it sold OK for our small niche products

but when we wanted to build a 2nd batch, a year or so later, the parts were unavailable in distribution, with official 18-24 mo lead time

I think if you buy something standard like max232 from them they are ok, but for the special parts
it seems like they make pizza mask prototype run so you can get samples but unless some one orders
a million they never get produced in volume

[jakeisprobably]

Iwould love to kick whoever writes the datasheets/appnotes (or who decided the policy on how to write those) for st parts square in the nuts

   I have seen a few comments like this elsewhere as well. What is so bad about ST?
     I am very inexperienced, but have been trying to broaden my prospective as much as possible. I started a new datasheet manufacturers thread on here to help me get better acquainted with the full scope of the EE market.
   I have been researching ST off and on for 2 days in my spare time and they appear to have more information and informative assistance than any other manufacturer I have researched thus far. I've actually been overwhelmed by the sheer scope of it.
  Although just a minor note, I've also noticed their IC's always appear to be of high visual quality. Just comparing the 50 to 100 circuit boards I have in my parts salvage pile, ST's finish quality and consistency is outstanding if not the best. After seeing their direct sale microcontroller development environment, reasonable hobby level price point, and Dave's positive feedback on them, I'm practically a wannabe "fanboy" already. I think it's the only microcontroller brand out there that Dave hasn't condemned the user base to "Fanboy" status yet...I always wanted to lead a movement....
   But seriously what is the flip side of ST? What am I missing from what I can see on the surface?

[Jeroen3]

Thanks for the warning about maxim. I was about to use some of their specialties for a long tern availability project. (10 years)
Do they have public longevity dates and extendive pcn rapports such as NXP and ST? I can't find them.

[JPortici]

The problem with ST? It comes when you have to actually work with it. a lot of information is implied or miswritten.
Case in point, try to set up the SPI in an stm32f1. The set up procedure in the datasheet doesn't state that you need to tell the peripheral not to work in i2s mode... and nothing works. but if you look at the code generated by the peripheral library you'll see that another unmentioned register is set.
or maybe it IS mentioned but not where it should be, in the set up procedure, where seone who already uses micro only care to look at

Then you have code examples that doesn't compile for unknown errors (the fucking define has been defined)

this morning, i have do design a datalogger addon for a product with a small memory chip so i say myself hey let's look at stm micros to do that so we can start fiddle with them for something. The rtc calendar datasheet never explicitly mention the lenght of the record for the date. It only says 'numbers are saved as BCD'. But how many for the year? and the sub second?
Shit like that.. i think i'll use a pic (or maybe a psoc) for that

It might be me, that i'm so picky because i grew up with microchip datasheets. Take any microchip datasheet and look at any register definition. Do the same in a stm datasheet. That will be pretty self evident (at least, for me, it is)

[poorchava]

Maxim for Monstrous lead times and high prices.

Microchip for crappy,  inefficient and expensive (relatively) microcontroller ridden with bugs. Their non-mcu stuff is mostly ok though.

Sent from my HTC One M8s using Tapatalk.

[Kjelt]

Maxim can be ok in my experience if you deal directly with them and make large contracts.
If you want to use their ics for small production runs (hundreds to thousands) you better make sure you have those ics in your factory stock, because indeed they might be hard to get.
But is that not the case for all special ics, I mean if you want to run production you need to know the lead times and reserve/order in advance. Couple of years back with the tsunami I remember that japanese elcos had a 18 month lead time and some jelly bean parts were also unobtanium.

[T3sl4co1l]

Spotted an odd one today,
https://www.fairchildsemi.com/datasheets/FP/FPF2280.pdf
Note the description "low off-state current <1uA".

I don't like descriptions.  Alright, let's check the tables. . . .huh...

Tim

[Cerebus]

Spotted an odd one today,
https://www.fairchildsemi.com/datasheets/FP/FPF2280.pdf
Note the description "low off-state current <1uA".

I don't like descriptions.  Alright, let's check the tables. . . .huh...

Tim

If your "huh" is at the Quiescent current of 58/100uA, perhaps you didn't spot the 'condition' column which has #EN=0V, i.e. not off-state. Or perhaps it's at the mention of it in what I tend to think of as the marketing claims section of the data sheet but its conspicuous absence from the more substantive section of the data sheet. One parameter that always seems to go this way is 1/f noise corner frequencies - lovely spec in the up front blurb but no mention in tables and no limits or ranges on the noise graph, if there is one.

Generally I find Fairchild data sheets OK to good, but I have to agree that this isn't one of their finest. I'd hazard a guess that this is one of those parameters that they know the part will meet if some other parameter passes testing so they don't test it and if they don't test it they don't list it in the characteristics table.

I have a general peeve with most data sheets that they don't make their conventions clear. e.g. The above example - does inclusion in the characteristics table, without a qualifying footnote, mean that a parameter is tested? There's a tendency to not be clear about what is tested, what is known only by characterization and what is "guaranteed" by design. Usually the figure that most matters to your design is the one that is most vaguely specified.

[T3sl4co1l]

Yes indeedy -- they give I_Q active (\EN low), but not inactive (\EN high)!

Tim

[Jeroen3]

The problem with ST? It comes when you have to actually work with it. a lot of information is implied or miswritten.

You have to shift standard identifiers in a CAN filter. Little details that were not written at first.

[ovnr]

ST has had some datasheets with incorrect information (no, not microcontrollers), so I generally avoid their parts if they're not just second-source jellybean stuff.

[tszaboo]

Maxim because they are Maxim
AMD because their embedded stuff became NRND before released
Intel because they strategically screw up everyting in my area of interest (I will buy it in my PC of course)
Renesas, Qualcomm, Broadcom, Mediatek, Sony because NDA and MOQ, or they just dont make anything interesting
Infineon, becasuse their microcontrollers suck big time
Microchip most of the time, because the Errata breaks your design all the time. There are big exceptions with this one though.
Anything made by Chinese companies.
Cypress semiconductor because they are weird

And the list goes on. I think it is easier to say, who I like.

TI, AD, Linear, ST, NXP, Fairchild, ON, Atmel, FTDI (yes), Freescale, IRF, IXYS.
You can design almost everything with them. No need to go to the wild west.

[bson]

think i'll use a pic (or maybe a psoc) for that

NXP has much better documentation than ST.  They also tend to have peripherals that are easier to work with from a software perspective.  ST for example loves 16-bit timers, with no obvious means of chaining them, while NXP parts tend to use 32-bit timers with one or two 16-bit ones for special uses.  If you've ever had to chain timers in an interrupt handler in a real-time system, say to implement a deadline scheduler, you know how much pain this introduces.  Especially in Cortex parts where the FIR is no more.  Other examples include weird-ass CSR flag management that's both poorly documented and very complicated to get water-tight except in the most naive designs.  (Or accept the firmware might flake out and glitch once in a while.)  TI documentation is also very good, though the peripherals are on par with ST, maybe slightly easier to work with but nothing like NXP.  Good luck for example getting interrupt-driven bi-directional I2C to work on an MSP430... whoever designed that peripheral should be taken out back and shot.