ESR tester

The one I picked up from ebay for $100 new (with, after asking, for a 110VAC, 60Hz supply -- added by the seller to adapt their 220VAC wallwart making it two wallwart things if you can believe it) provids Kelvin leads and tests at 100Hz,

120Hz, 1kHz, and 10kHz. So some do more and don't cost a great deal.

Jon

_

's

a

I've written tons of adsp assembler and once you have done enough fix point you it is not that hard, but for a lot of stuff it just isn't worth the effort

and yet,

stm32f407; FPU, 1M flash, 160K RAM, 168MHz, datasheet says ~46mA@3.3V

ADSP2181; no FPU, no flash, 80K RAM, 40MHz, datasheet says 85mA@5V

There's probably 15 years between them, but times change

-Lasse

If there's a distinct flat region in the Z vs frequency curve, that's the ESR zone.

In the pic I posted, things were nicely separated into three zones. That seems to be the case for pretty big electrolytics.

--

John Larkin         Highland Technology, Inc 

jlarkin at highlandtechnology dot com 
http://www.highlandtechnology.com 

Precision electronic instrumentation 
Picosecond-resolution Digital Delay and Pulse generators 
Custom laser drivers and controllers 
Photonics and fiberoptic TTL data links 
VME thermocouple, LVDT, synchro   acquisition and simulation

Yeah, the feature size change alone would account for all of that and more. Best would be to find and compare two products on the same process and feature size, one with and one without the FP part. There is no question that FP occupies die space, reduces yield therefore, and takes power (unless it can be isolated and turned off.) And usually is accompanied by more pins because when you add FP your market is different and more pins make more sense, each pin of which cost money which adds BOM cost....

The problem for a vendor is that the marketplace of programmers doesn't include a large number of numerically savvy folks who are versed in using, for example, a barrel shifter capable of leading bit recognition or in designing algorithms that work well with all these bits and pieces you see on an ADSP-21xx, for example. So the programmer market is small, forcing a customer to find them and pay them. The work is custom and doesn't port, as well. C compilers that can effectivelyu take advantage are another issue. Etc. So it's certainly not "low hanging fruit" for them and a manufacturer won't pursue it without some good reason, I guess.

But I like them a lot. I was able to substantially reduce the power consumption at the time, and cost, over the existing FP units available and this meant a decided advantage in marketing the product against competition who chose an "easier" path.

Jon

At least 15 years! It's apples and oranges. And I'd look really close at the STM32F407 part to be sure that 46mA happens while floating-point is happening (I ain't saying it's not -- just that you should look).

I'll bet that the equivalent Cortex M3 part uses less power when all of the (fewer) stops are pulled, and I'd be kinda surprised if the Cortex M4 parts don't use more power when you do floating-point calculations than when you refrain.

It's too bad they didn't make a Cortex "M3.5" with the DSP enhancements but no floating point -- but then, it seems like more and more the processor die space is getting to be less and less of the total chip space.

--
My liberal friends think I'm a conservative kook. 
My conservative friends think I'm a liberal kook. 
Why am I not happy that they have found common ground? 

Tim Wescott, Communications, Control, Circuits & Software 
http://www.wescottdesign.com

y t

se

sure it is apples and oranges, but shows that things are changing and it might very well be that for many things using floating point takes fewer instructions that getting everything aligned for fixed point so it come out ahead

And if you look a picture of the die, most of it is IO and huge blocks of memory, I'm not sure how much it really adds to have the FPU

-Lasse

The intel chips, like the slightly older Prescott in the link below, are probably the more extreme case where FP takes up a smaller proportion of die space. EVERY microcontroller that has FP will have a larger percentage of their die used by FP, I believe, without exception. This is, in part, because of the huge L1 and L2 caches they've stuffed ondie, plus the parallel instruction decoder and out of order execution and branch prediction they've added.

There you can see the floating point occupies the right lower corner plus the floating point registration station and scheduling seen in the upper right area. Note that it is still quite significant, even in this unusual case where there is so many other significant functional units in play.

Microcontroller die maps, if you can find them available, will certainly show a higher percentage IMO. Even today. It was much worse 20-30 years ago when the other functional units were significantly smaller and L2 caches, for example, were on a separate die connected via the backside bus (in the intel case of, say, the PPro and P II.) If you find a die map for the ARM VFPv2 or VFPv3, for example, I'd love to see it in context. Didn't find one readily available, though. Same for the MIPS R14k -- no map easily found... but I'd love to see one.

Jon

re

nly

ny

it.

s a

ee.

ctly

ite

x

t at

V t f

M4

n s

I'd think the Intels are different from MCU, the have tons of registers, pipelining and all kinds of trickery to make it run at GHz

look at this one,

nearly half of it is memory and it is only 16K flash + 4K ram, so unless they use the same die for all they parts, the ones with bigger memory will be almost all memory

-Lasse

I can't tell anything from that about floating point support. Can you?

No question about it. When you have more transmission gates and inverters than you know what to do with, wasting them on cache and memory always comes to mind. Lots of folks do it.

Jon

tware

s only

o any

or it.

less a

e

agree.

exactly

Quite

fix

't

g it at

r

or

3.3V V

se

oint

.

l of

tex M4

than

ents

p s

you can look up the part number ;)

It's a Cortex M3 it doesn't have FPU

The point was that if you look at the die of modern MCU even with out FPU it looks like a memory chip with a cpu in the corner

I wouldn't want to go back to MCUs that needs external RAM and FLASH

If you can fit what you need on chip it is easier, faster, more reliable and uses less power

-Lasse

software

only

any

it.

a

agree.

exactly

Quite

at

M4

You misunderstand. I mean I can't tell about the AREA occupied by the FP and it's support circuits. I CAN do so with the link I provided on the Prescott.

I still would still need to see the map all lined out. I provided a diagram of the Prescott, which is most certainly a case that makes my point even though the Prescott has so much else to do, as well, and includes substantial cache memory besides.

I guess I'm just not sure what you are saying. My point doesn't appear to be affected by anything you've said so far.

No one wants to go back to the 1980s micros.

Jon

e:

software

plus only

. So any

n for it.

le less a

the

of

of agree.

to

ng exactly

ay. Quite

ugh fix

isn't

oing it at

e's

ower

s for

..@3.3V

A@5V

close

g-point

ok).

all of

Cortex M4

ns than

cements

he

chip

and

kes

ocks

My point is that if the chip is 80% memory and 10% peripherals the impact of the the cpu core growing 20% or even 50% isn't much

-Lasse

If, then, yes. Problem is, I can't tell that from the diagram in that case or any case so far presented. Even in the Prescott case where there is some descriptive text you don't see that ratio.

I easily admit it is easy to throw lots of memory at a chip die to wash out the impacts of other sections, in this day when there are so many transmission gates and inverters. Was less easy to do 25 years ago, though. And until I see a die nicely mapped out, at least as well as the one I presented to you, I remain unsure about any arbitrary 10% figure you may throw out today.

Jon

I think that's the rub. My DC Load doesn't have NEARLY the resolution or sampling time to make these kinds of measurements.

So. What make and model?

?-)

Tonghui TH2821A. From China. If you find yourself getting something like this overseas and want to use it with US

120VAC, be certain to email the supplier about your circumstances. I made certain to do so, just dotting i's and crossing t's just in case, and it turned out that I am darned glad I did because the response I got back was, "Uh... okay... will need a few days to locate an adapter... is that okay?"

Turns out, they shipped the original unit designed for their

220 supply, 50Hz system. The wallwart was a small affair -- 220V, 50Hz, 40mA source to 12V@150mA output. They then also added, at no extra cost to me, a monstrous adapter -- an HX-203A -- designed for 110-120VAC, 50/60Hz input and to provide 220-240VAC output at 100 watts! The other wallwart plugs directly into it.

Works. And what the heck... I've now got a unique adapter that I didn't have before, too. ;)

Jon

CPU

this.

the

Ah yes. Reminds me of my test engineer, days 20+ years ago. There was the calibration service unit and the primary standards lab. Sometimes we had test reference devices measured annually by the primary standards lab to verify both our operation and the in house calibration service. You might imagine the foofuraw that erupted when one of the verification devices exhibited a dramatic change.

?-)

That is a good supplier. And now will probably get more business from your report. I recently bought a 27 inch superhd (2560 x 1440 resolution) LCD monitor and when it came in with a power cord set for the country of origin (Taiwan, 220 V 50 Hz) i contacted the manufacturer about using it with 120 V 60 Hz power and could they supply a power cord. They answered it will work fine, but they did not have the means to supply the power cord. So i grabbed one of my extras and started using the monitor, works just fine.

?-)