Fast PNPs--suggestions?

specified

That is a dicey game with discrete BJT.

Coulda told us that earlier :-)

Panasonic used to have the best but AFAICT they have discontinued just about all that I checked. Same with Sanyo :-(

These look good but 0.65pF:

I don't know how well a BF550 would fare because its datasheet is grossly incomplete.

You could contact these guys but it would probably only work if you can hitch a ride on someone else's production run:

specified

I'm not really Joerg, I just play him on TV. (All bald white guys look alike.) ;)

I'm using the ARM because (a) it's pretty cheap, (2) it has reasonable-for-a-MCU quality 12-bit ADC and 2 12-bit DACs, and (3) I don't have time or energy to invest in writing libraries for more than one family, and ARMs come in an enormous range of cost and computing power.

So an extra buck on the CPU is a good deal in this case, I think.

Cheers

Phil Hobbs

nothing specified

capacitance <~

...and if Obama puts your supplier out of business, you can switch to another without too much lost sleep.

For a small company building small quantities, there is no other rational choice, IMO.

specified

Sure, but if anybody could do it, they wouldn't need to license it from me. ;)

DS shows 325 MHz typical f_T at 10V, 1 mA, which is a factor of 10 too low, unfortunately.

Thanks, I'll check them out. A unique device can sometimes give a fairly serious competitive advantage, but only if the last drops of juice have been squeezed out of the circuit topology first. I'll need to breadboard the basic noise canceller with these PDs and make sure.

The thing I was most worried about is actually the speed variation of the PDs across their faces. Thanks to a little diode laser driven by an SD-24 TDR pulser, I find that at sufficiently large bias, the PDs I'm using have less than 30 ps rise time or skew variation anywhere on the die. That's accurate enough that I'll get good cancellation regardless of beam position, which is key.

Cheers

Phil Hobbs

specified

But no form-fit-fucntion 2nd sourcing, right?

Sure, the 80C51. No ADC but about the only uC I know of that has 2nd sourcing. Plus the clients can find a programmer pretty much everywhere.

specified

capacitance<~

Using the hardware abstraction library (CMSIS), different Cortex micros are code-compatible with a recompile. Porting the code to a different manufacturer's M3, or even a different Cortex model is a lot simpler that way, or so I'm told.

Simon got his data acq system running with AVR Studio 5, using hardware he built himself. Now he's working on the beginnings of the gizmo-tweaking library using the STM32 Discovery board and a crippleware version of IAR Embedded Workbench. (Somebody at IAR has a sense of humour--I gather that their standard file extension stands for Embedded Workbench Workspace, i.e. ".eww". )

There are a bunch of STM devices that are upward-compatible with the STM32F100x, so they'll be around for awhile, and moving to another vendor will probably just need a board spin. This particular device is enough for us to deal with at the moment!

Cheers

Phil Hobbs

specified

Package might change (form) but as long as you plan ahead, everything else should work. A board spin and you're off and running.

For *really* low function, maybe. Maybe. For any complication, ARM has it all over everyone. It's *everywhere*.

nothing specified

capacitance <~

^^^^^^^^^^^^

That is very much frowned upon in industry. New validation runs, regressions tests, new EMC runs, a whole fresh ECO process, inventory issues, and so on. If this happens shortly before the end of a quarter and the inventory numbers surge a bit because of it the CFO ain't gonna be happy at all.

You'd be surprised. Even our pellet stove has an 8051 in there.

or

nothing specified

capacitance <~

That's the worst case. What's your alternative? A bottoms-up redesign? ...including code and verification? Ick!

another

I trust your pellet stove isn't very complicated (i.e. "*really* low function"). A boards spin, in the event a CPU manufacturer bit the dust, wouldn't be much work, either.

FT92 or

nothing specified

tput

citance <~

my

pay not

and

from memory, 4, or is that 8?, 16 bit ADC's parbly not fast enough, though

or

nothing specified

capacitance <~

No, I like to stick with the 89C51 architecture. Then you can have multiple sources. Meaning no new layout, no new EMC cert, no inventory issues. You use up the current uC stock and then switch. Even the ECO is simple because you validate the compatibility and only release another vendor for the same part. That is a huge difference in NRE for the business.

I have one design with a 89C51 in there that is now in its 17th year of production. No end in sight, it's alomost like the VW Beetle. I don't think that would have been possible with any other architecture.

another

In medical, aerospace and other such regulated fields, it is a ton of work. All the stuff described above and no shortcuts allowed.

BFT92 or

nothing specified

capacitance <~

not

and

I don't mind the x51 for projects where a trivial CPU works. I don't get paid for such things often, though.

inventory

What do you do when an x51 is about three or five orders of magnitude too slow to be of use?

...and your point is?

another

Apparently you're not bothering to actually, you know, read what I've written.

You mean a GBT? ;-)

Tim

They'd rather go out of business than do the paperwork for two new parts?

BFT92 or

nothing specified

side--output

capacitance <~

my

not

and

no inventory

I am an analog guy so don't get involved in that much. But sometimes I do and then I usually recommend a DSP. One that is used in industrial or avionics because those tend to be around for a long time. But not as long as the 51 does.

My point is that my client was able to continually produce this board without any redesign due to parts obsolescence. Now try that with any other uC.

another

You wrote, and I quote "A boards spin, in the event a CPU manufacturer bit the dust, wouldn't be much work, either".

That is not correct. It is a lot of work. Also for a stove manufacturer because they do not have in-house design teams. They have to go out and line up a contract design. That costs tons of money and requires a lot of hand-holding by the project manager. I know that because I get involved in such processes on a very regular basis.

BFT92 or

nothing specified

side--output

capacitance <~

my

not

and

no inventory

Therefor there they are of no use?

DSPs aren't suitable for all tasks, and neither are 8051s. The ARM is a perfect answer to a majority of these applications. Some argue, with good reason, that the 8051 isn't an answer to any.

...and what do you do when your application isn't trivial? Oh, that's right, those applications don't exist, since that isn't your thing.

another

rational

it

Right. Now WTF are you going to do with a non-trivial application? Ignore it?

That's a TRIVIAL application. Good grief!

BFT92 or

nothing specified

side--output

capacitance <~

started my

that I

pay not

calibration and

look

else

paid

no inventory

slow

To a degree they are, which is why I have some exposure.

Then those "some" are wrong :-)

There's reasons why Cypress picked the 8051 architecture for many of their PSoCs.

As I wrote before, then we usually go the DSP route. Plenty of horsepower there. I have nothing against other higher-powered uC architectures. I just do not advise using them where an 8051 suffices. Because they lack 2nd souring.

another

rational

it

written.

See above.

What difference does it make? Do you think they do not have to go through validation? What would you advise they do about EMC? Whistle innocently and not do a re-test? What if a competitor snitches about that? Happens a lot more often than you might think.

Taking this lightly as in "Oh, it's totally trivial!" and not testing is a quick path to a nasty recall situation. We just had one on stuff that we purchased, 1.6 million units had to be fixed. Sure looks like it was because of lack of testing. That is majorly painful.

the

BFT92 or

nothing specified

side--output

capacitance <~

is 0.5p

started my

that I

pay not

calibration and

look

than

else

paid

cert, no inventory

business.

slow

Huh? Because you don't do anything non-trivial, non-trivial applications don't exist? Narcisistic, much?

No, there are good reasons. I'm not so much of an x51 phobe, but there are good reasons not to use it.

...and ARMs for the rest. They must be really stupid, too.

Again, because that's your application base, noting else exists? You keep repeating the same thing but I guess I'm having a hard time believing that you're that narrow.

another

rational

everywhere.

has it

written.

What you wrote above says that you do. ...or that if you don't do it, it doesn't exist. Really weird.

This is getting ridiculous! What do you do if your application is *NOT* trivial. Some of us have such jobs.

Good grief. I guess you just don't want to read.

the

BFT92 or

milliamps, nothing specified

side--output

capacitance <~

is 0.5p

started my

that I

pay not

calibration and

look

than

else

paid

cert, no inventory

business.

slow

I did not say that.

Sure, a MSP430 is nicer to program, has a 16-bit engine, and I like its architecture better than the 8051. But, no 2nd source.

No, because there is a market for both.

I have repeated the reason why the 8051 architecture is best in many places, whether you accept that as a reason or not doesn't really matter because it is still a fact: Second sources.

to another

rational

everywhere.

has it

written.

Where have I written something doesn't exist? Quote please.

Don't jump back and forth here. You said the stove application is trivial. I replied that it makes no difference as to the effort of re-spinning a board. If the application is not trivial it is also a lot of effort.

In an app that in your eyes is trivial you can avoid such hassles by using a 2nd sourceable 8051 type. If you don't want to do that then that's your business.

In an app that is not so trivial you still have choices. You can either use a fancy uC that possibly falls from grace five years down the road and have an unhappy boss, or you can use a DSP that is not such a nice fit but is likely going to be around 10 or more years. My preference is generally the 2nd path.