Voltage to PWM chip (similar to class D)?

On Sun, 01 Jun 2014 03:50:11 -0400, rickman <gnuarm@gmail.com> wrote:

I pretty much only used the 8051 when I was allowed to decide. Clients
have used Atmels and also PICs that have lasted a long time but I don't
have production data for those products.

Yes, some MCUs live for a while, but not 20 years. I have never seen an
MCU other (than the 8051 and a few Mil/aerospace processors) that long.
Why? Because no one wants them. They are old, slow and not cost
effective. If you seriously need a 20+ year production life without
redesign of any sort, then you will need to stick with 8051s and 4000
series CMOS. Is redesign really that big a problem that it can't be
done every 10 years or so?

If you are designed a product for multiple decades and expect to do
some redesign every 10 years, you really need to be careful with the
initial design. The product should be very modular with well defined
(vendor independent) interfaces. If you need to redesign a module for
whatever reason (availability/cost reduction), it should not be that
hard to redesign a single module (card) without affecting the rest of
the product.

If the product is also using some hierarchical structure with high and
low level interfaces, in the next revision with high performance
components to replace .g. a rack of cards with a single card, possibly
also moving some HW functionality to SW. As long as the original rack
level interface is implemented on the new, highly integrated card
level module, things should be pretty easy to handle.

I am pretty sure they will supply. They told me so and they've never
lied to me. Their prices are on the high side anyhow and I don't think
they would jack up the price on customers.

I have no idea what you base that on. If you ask them to make a part
they have discontinued it will cost an arm and a leg. Just give it a
thought for a bit. Running a fab is not cheap and making a run of parts
is not done for less than how many thousands of chips?

Companies come and go quite quickly, some become uncompetitive and go
bankrupt, some are taken over by a hostile competitor.

There are political risks as seen in Eastern Europe and Asia in the
last two decades. Also trade embargoes (such as CoCom) may limit into
what countries you can sell your products, if you are using components
originated from specific countries.

 

[Joerg]

Lasse Langwadt Christensen wrote:

Den søndag den 1. juni 2014 02.16.57 UTC+2 skrev Joerg:
Lasse Langwadt Christensen wrote:

Den l�rdag den 31. maj 2014 16.21.47 UTC+2 skrev Joerg:
rickman wrote:
On 5/30/2014 7:59 PM, Joerg wrote:
Stopband is a pretty gradual thing when you can't go past 3rd order on
the output filter. This application is sensitive to noise because the
load is super fast.
I guess I'm not really following all this. You have a control circuit
that is not easy to do with analog chips and yet you don't want to use a
digital control. From the bits and pieces I am seeing I believe a
classic ADC-DSP-DAC would suit this problem very well. ...
Generating a PWM stream with down to 1nsec granularity that way would be
increadibly power intense and expensive.
the dsPIC33 has ~1ns resolution PWM and a 10 bit 2Msps ADC, $1.69 if you buy 1000


How does it do that when the on-chip VCO for system frequency is spec'd

between 100MHz and 200MHz?


never used it so I don't know, look like you have to set it for 120Mhz to get 1.04ns resolution

That would only be possible with silicon delay lines in there.

Silicon delay wouldn't be so nice.


why not?

My experience with silicon delay lines is that they have a lot of phase
noise. I've ripped several of them out and replaced them with
controllable LC circuitry.

120MHz clock does the coarse duty cycle, a string 8 delays of nominal
1.04ns, servo the delays to make up exactly one cycle and you have 8
phases to do the fine setting duty cycle

I guess if the VCO is done with ring oscillator there is already a
string you can tap to get the extra phases for free

I believe Xilinx does something similar servoing of delays for their
IO delays

A ring oscillator could work although phase noise is often not so
stellar either.

Anyhow, with so much effort and large CPU chips I might just do the
analog thing.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

rickman wrote:

On 5/31/2014 10:16 AM, Joerg wrote:

It was a cutting-edge FPGA in the mid 90's. With ASICs you can get much
more design security. The trick is to pick a foundry you trust and a
run-of-the-mills process that is used for tons of other products. By
going directly to the foundry you are cutting out one middleman (the
FPGA vendor) and thus reduce the overall risk.

I don't know what a "cutting edge" FPGA is other than a startup company.
Is that risky... yes of course.

With cutting edge I meant the latest and gratest (newest) from a major
FPGA vendor. The usual, more gates, faster.

You are speculating about the foundry vs. FPGA. All foundries are used
for "tons" of other products or they wouldn't stay in business. In
fact, the FPGA companies are the customers that define the foundry
business being the first real customers of the state of the art new fab
lines.

FPGAs are among the longest lived complex semiconductor devices out
there. You may find transitors, op amps and 8051s that live a lot
longer, but you won't find much LSI that lives longer than FPGAs.

Lots of the logic chips have been around since my childhood days and no
end in sight. For example, almost the whole CD4000 series which I gladly
use in designs, for that very reason.

An example for the longevity of semiconductor processes: We just started
something on a 4" wafer. Those date back almost to the days of
Methusaleh yet it's no problem. If a process is also used to make mil
stuff, chances are it'll survive both of us.

So? You can't find anything that competes with FPGAs in that technology.

Many analog/mixed things do not need FPGA. They compete differently.
Like a fast pulse-echo thing I just finished. It could be done in FPGA
but it's all analog, in this case for cost reasons. It is hard to beat
3-cent transistors.

And I might. Plus opamps/comparators.

Ok, but you are here because you are finding this approach to be a PITA
because of all the accuracy issues.

I can get the accuracy alright. I am here because I couln't imagine
there not being a class-D modulator with DC accuracy. But it seems there
truly isn't one.

I pretty much only used the 8051 when I was allowed to decide. Clients
have used Atmels and also PICs that have lasted a long time but I don't
have production data for those products.

Yes, some MCUs live for a while, but not 20 years. I have never seen an
MCU other (than the 8051 and a few Mil/aerospace processors) that long.
Why? Because no one wants them. ...

Huh? I have many clients who want exactly that. They want a product they
ideally won't have to touch and can just keep selling. This is one of
the reasons why products such as pellet stoves and other reidential and
commercial gear use the 8051.

... They are old, slow and not cost
effective. If you seriously need a 20+ year production life without
redesign of any sort, then you will need to stick with 8051s and 4000
series CMOS. Is redesign really that big a problem that it can't be
done every 10 years or so?

In many cases, yes, it is undesirable.

I am pretty sure they will supply. They told me so and they've never
lied to me. Their prices are on the high side anyhow and I don't think
they would jack up the price on customers.

I have no idea what you base that on. ...

On trust. I have developed a trust level with that company over decades
and (in contrast to some others) they have never disappointed.

... If you ask them to make a part
they have discontinued it will cost an arm and a leg. ...

Again, they call first and they won't discontinue if there is still a
product line that uses the chip. I assume they'd expect reasonable
quantities but that is generally the case with my designs. So it's not a
question of re-starting an old IC, it is one of not stopping it in the
first place.

... Just give it a
thought for a bit. Running a fab is not cheap and making a run of parts
is not done for less than how many thousands of chips?

We are right now doing a run of about 10,000 or so, depending on the
yield. It's not that bad if you spread the cost over each resulting device.

Most ASICs I have been involved in run between 100,000 and a million
parts per year which in the world of semiconductors is a drop in the
bucket. They are all in the low single-digit Dollar range. Order gets
called in, they make another round of wafers. Often they pre-make them
and just dice them up after the order comes in. So you have to give them
a reasonable forecast. It's almost like running a catering business.

They _are_ making 100MHz 8051. No kidding. If it ain't got enough horses
use a 2nd one

And is that multiple sourced and long lived.. I doubt it. Even so, 100
MHz 8051 won't do this job without some analog support which takes you
back to your accuracy problems.

That's true, which is why I'll probably do it analog this time around.
When using 0402 sizes I can get it onto the real estate similar to a
micro controller. I was hoping to make it smaller but it seems that's
not in the cards.

This is one reason why I try to keep things simple. Many jobs that are
done with a DSP can be done without. But not all of them, of course. Of
course, for me that's easier to say than probably for you because I do
mostly analog stuff. Sometimes it's the whole architecture though and
then one of my fuirst questions to the client is about parts sourcuiing
and longevity.

Who was it that said things should be done as simply as possible, but no
further. You are here because the analog approach is not so simple due
to the... well, analog effects.

Not quite. I can do it analog and with high precision. It's just that I
was hoping there'd be a solution in a can, in the form of a li'l class D
chip that happens to sport a good DC accuracy. That's why I am here, not
because an analog solution won't work. More because a good analog
solution consumes real estate but so does a processors with clock
crystal and all.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

josephkk wrote:

On Sat, 31 May 2014 07:23:14 -0700, Joerg <invalid@invalid.invalid> wrote:

I am going to do something like that (but probably not with a 555) if
nothing single-chip comes up. That's the reason for this thread, to see
if there isn't anything out there. I mean, every class-D amp must have a
super-linear PWM generator. It's just that most have the power stages
built in (would be ok, can be left idle) and have lousy or no DC
performance (would not be ok).

Ok a weird question: Does the response need to be 0.1% for any step size
at 15 kHz? ...

Nope :-(


... Can it be "slew rate limited" a bit to get to 0.1% for a large
step; to like 900 Hz?

Unfortunately that would put a serious crimp into the versatility of the
product.


OK. Instead of a strict linearity perspective can we look at it from a
settling time perspective for various portions of step size from minimal
to full scale?

That's where the bandwidth spec comes in, essentially 15kHz. So an audio
chip would be very nice. But it seems there isn't one.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

josephkk wrote:

On Sat, 31 May 2014 07:25:08 -0700, Joerg <invalid@invalid.invalid> wrote:

josephkk wrote:
On Thu, 29 May 2014 14:20:26 -0700, Joerg <invalid@invalid.invalid> wrote:

Folks,

Does anyone know an IC that can turn a control voltage into PWM and can
handle PWM frequencies in the 50-1000kHz range? Similar to a class D
driver but has to go down to DC. The changes in control would be
restricted to the audio spectrum below 15kHz.

The LTC6992 does this nicely but isn't precise enough. Same with
555-style timers or switcher chips. I am looking for better 1% and
ideally a lot better, including nonlinearity, drift, warts and all. A uC
is not suitable either because it should be simple and I need very fine
control granularity, down to around 0.1%.

Can't use short-lived consumer chips for radios and TV sets and such.
The only things that i have seen that are even close were V to F
converters. And they did DC to a few hundred Hz on the input spectrum.

Yup, I've used them but they aren't precise enough. Also, they don't
work well for PWM, just for V/F anf F/V.

Does it HAVE to be PWM? Can't PFM do the job?

Unfortunately not in this case.

Maybe some audiophool Class D amplifier IC.

T'is exactly what I am looking for and why I posted here

I had noticed.

Yeah, and since it seems there isn't one I thing it'll be the old analog
nose to the grindstone job again.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

Chris Jones wrote:

On 30/05/2014 07:20, Joerg wrote:
Folks,

Does anyone know an IC that can turn a control voltage into PWM and can
handle PWM frequencies in the 50-1000kHz range? Similar to a class D
driver but has to go down to DC. The changes in control would be
restricted to the audio spectrum below 15kHz.

The LTC6992 does this nicely but isn't precise enough. Same with
555-style timers or switcher chips. I am looking for better 1% and
ideally a lot better, including nonlinearity, drift, warts and all. A uC
is not suitable either because it should be simple and I need very fine
control granularity, down to around 0.1%.

Can't use short-lived consumer chips for radios and TV sets and such.


Maybe you don't strictly need PWM, but would be able to use sigma-delta
a modulated bitstream instead. In that case you could perhaps use this
one, though it comes with galvanic isolation that you might not need:
http://www.analog.com/static/imported-files/data_sheets/AD7401A.pdf
Perhaps there is something similar and cheaper without the isolation.

We had discussed that earlier in the thread and unfortunately it won't
work. Has to be clean PWM.

Otherwise if you really need PWM, what if you wrap feedback around the
LTC6992, i.e. invert and square up its output by putting the PWM through
a CMOS inverter running from a very stable power supply, and then use
resistors to add the inverted PWM waveform to the incoming input
voltage, and call the sum the "error voltage". Theoretically the average
of this "error voltage" will be exactly mid-rail. Use an op-amp to
integrate any deviations from mid-rail and feed that into the LTC6992,
to correct it. Perhaps two op-amps would be needed as integrating
circuits tend to invert, which you might not want here.

That's similar to what I was going to do. But I wanted to avoid the
extra parts for real estate reasons, hence my thread here inquiring for
a fully integrated solution. Class D has super clean PWM but I (so far)
don't know any that are DC-accurate.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

Joerg wrote:

Chris Jones wrote:
On 30/05/2014 07:20, Joerg wrote:
Folks,

Does anyone know an IC that can turn a control voltage into PWM and can
handle PWM frequencies in the 50-1000kHz range? Similar to a class D
driver but has to go down to DC. The changes in control would be
restricted to the audio spectrum below 15kHz.

The LTC6992 does this nicely but isn't precise enough. Same with
555-style timers or switcher chips. I am looking for better 1% and
ideally a lot better, including nonlinearity, drift, warts and all. A uC
is not suitable either because it should be simple and I need very fine
control granularity, down to around 0.1%.

Can't use short-lived consumer chips for radios and TV sets and such.

Maybe you don't strictly need PWM, but would be able to use sigma-delta
a modulated bitstream instead. In that case you could perhaps use this
one, though it comes with galvanic isolation that you might not need:
http://www.analog.com/static/imported-files/data_sheets/AD7401A.pdf
Perhaps there is something similar and cheaper without the isolation.


We had discussed that earlier in the thread and unfortunately it won't
work. Has to be clean PWM.


Otherwise if you really need PWM, what if you wrap feedback around the
LTC6992, i.e. invert and square up its output by putting the PWM through
a CMOS inverter running from a very stable power supply, and then use
resistors to add the inverted PWM waveform to the incoming input
voltage, and call the sum the "error voltage". Theoretically the average
of this "error voltage" will be exactly mid-rail. Use an op-amp to
integrate any deviations from mid-rail and feed that into the LTC6992,
to correct it. Perhaps two op-amps would be needed as integrating
circuits tend to invert, which you might not want here.


That's similar to what I was going to do. But I wanted to avoid the
extra parts for real estate reasons, hence my thread here inquiring for
a fully integrated solution. Class D has super clean PWM but I (so far)
don't know any that are DC-accurate.

What I meant is just a plain old feedback loop. Demodulate the PWM via a
simple lowpass and us that as the error signal.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Steve Wilson]

Joerg <invalid@invalid.invalid> wrote:

Many analog/mixed things do not need FPGA. They compete differently.
Like a fast pulse-echo thing I just finished. It could be done in FPGA
but it's all analog, in this case for cost reasons. It is hard to beat
3-cent transistors.

I know you like to roll your own, but what value do you assign to system
design, board stuffing, checking for proper solder joints, on-board
diagnostics, troubleshooting, repair, documentation, procurement, incoming
inspection, etc.? It would seem a 3 cent transistor would end up costing
much more, but the question is how much more? And how do you calculate it?

The whole issue probably belongs in the BOM Minimization thread

 

[rickman]

On 6/1/2014 11:09 AM, Joerg wrote:

rickman wrote:
On 5/31/2014 10:16 AM, Joerg wrote:

It was a cutting-edge FPGA in the mid 90's. With ASICs you can get much
more design security. The trick is to pick a foundry you trust and a
run-of-the-mills process that is used for tons of other products. By
going directly to the foundry you are cutting out one middleman (the
FPGA vendor) and thus reduce the overall risk.

I don't know what a "cutting edge" FPGA is other than a startup company.
Is that risky... yes of course.


With cutting edge I meant the latest and gratest (newest) from a major
FPGA vendor. The usual, more gates, faster.

I don't know why you can't be specific about the vendor and part. I
can't think of an example where one of the FPGA companies obsoleted a
part any time within a 10 year time frame. I have a recollection of
Xilinx dropping a package/part combo a bit earlier, but I'm not sure.

If you can go the ASIC route, then by all means do it. But FPGAs are
usually used where ASICs are cost prohibitive because of the low volume.

I remember when some of the Xilinx reps would post in c.a.fpga about how
the ASIC market would continue to shrink to nearly nothing because of
the high mask costs. But they were not comparing apples to oranges...
which they should have been. If you need today's technology in an FPGA
you can do the design in a 4 year old ASIC at least.

You are speculating about the foundry vs. FPGA. All foundries are used
for "tons" of other products or they wouldn't stay in business. In
fact, the FPGA companies are the customers that define the foundry
business being the first real customers of the state of the art new fab
lines.

FPGAs are among the longest lived complex semiconductor devices out
there. You may find transitors, op amps and 8051s that live a lot
longer, but you won't find much LSI that lives longer than FPGAs.


Lots of the logic chips have been around since my childhood days and no
end in sight. For example, almost the whole CD4000 series which I gladly
use in designs, for that very reason.

Yes, we have already mentioned that. But you can't do much with SSI
logic. It is *no* comparison to LSI and VLSI. It is silly to even talk
about them in the same discussion.

An example for the longevity of semiconductor processes: We just started
something on a 4" wafer. Those date back almost to the days of
Methusaleh yet it's no problem. If a process is also used to make mil
stuff, chances are it'll survive both of us.

So? You can't find anything that competes with FPGAs in that technology.


Many analog/mixed things do not need FPGA. They compete differently.
Like a fast pulse-echo thing I just finished. It could be done in FPGA
but it's all analog, in this case for cost reasons. It is hard to beat
3-cent transistors.

Yes and you are here looking for a solution that you are finding
difficult in analog.

And I might. Plus opamps/comparators.

Ok, but you are here because you are finding this approach to be a PITA
because of all the accuracy issues.


I can get the accuracy alright. I am here because I couln't imagine
there not being a class-D modulator with DC accuracy. But it seems there
truly isn't one.

You seem to be contradicting yourself. I don't follow.

I pretty much only used the 8051 when I was allowed to decide. Clients
have used Atmels and also PICs that have lasted a long time but I don't
have production data for those products.

Yes, some MCUs live for a while, but not 20 years. I have never seen an
MCU other (than the 8051 and a few Mil/aerospace processors) that long.
Why? Because no one wants them. ...


Huh? I have many clients who want exactly that. They want a product they
ideally won't have to touch and can just keep selling. This is one of
the reasons why products such as pellet stoves and other reidential and
commercial gear use the 8051.

Yes, you have clients... but how many chips? Until the number reaches
millions per year no one cares and you are stuck with 8051s.

LOL, designing the 8051 into a pellet stove doesn't really mean a lot.
If the volumes were higher, they could save money going to the Asian 4
bitters. They are using the 8051 because they barely need an MCU. If
they need anything at all more complex they are stuck.

... They are old, slow and not cost
effective. If you seriously need a 20+ year production life without
redesign of any sort, then you will need to stick with 8051s and 4000
series CMOS. Is redesign really that big a problem that it can't be
done every 10 years or so?


In many cases, yes, it is undesirable.


I am pretty sure they will supply. They told me so and they've never
lied to me. Their prices are on the high side anyhow and I don't think
they would jack up the price on customers.

I have no idea what you base that on. ...


On trust. I have developed a trust level with that company over decades
and (in contrast to some others) they have never disappointed.

That means your needs have been compatible with their needs... so far.

... If you ask them to make a part
they have discontinued it will cost an arm and a leg. ...


Again, they call first and they won't discontinue if there is still a
product line that uses the chip. I assume they'd expect reasonable
quantities but that is generally the case with my designs. So it's not a
question of re-starting an old IC, it is one of not stopping it in the
first place.

I don't believe that for a minute. If you are the only customer left
they will keep the part in production? Really? Have they actually told
you that?

... Just give it a
thought for a bit. Running a fab is not cheap and making a run of parts
is not done for less than how many thousands of chips?


We are right now doing a run of about 10,000 or so, depending on the
yield. It's not that bad if you spread the cost over each resulting device.

Most ASICs I have been involved in run between 100,000 and a million
parts per year which in the world of semiconductors is a drop in the
bucket. They are all in the low single-digit Dollar range. Order gets
called in, they make another round of wafers. Often they pre-make them
and just dice them up after the order comes in. So you have to give them
a reasonable forecast. It's almost like running a catering business.

Now you are mixing apples and oranges again. ASICs are typically not
full custom. Your custom ASIC is shared across many customers with a
small number of custom layers. The ADI chips we were discussing are
full custom parts and running a batch is a big deal. I am sure if you
were the last customer for a part you would have to go pound sand.

They _are_ making 100MHz 8051. No kidding. If it ain't got enough horses
use a 2nd one

And is that multiple sourced and long lived.. I doubt it. Even so, 100
MHz 8051 won't do this job without some analog support which takes you
back to your accuracy problems.


That's true, which is why I'll probably do it analog this time around.
When using 0402 sizes I can get it onto the real estate similar to a
micro controller. I was hoping to make it smaller but it seems that's
not in the cards.


This is one reason why I try to keep things simple. Many jobs that are
done with a DSP can be done without. But not all of them, of course. Of
course, for me that's easier to say than probably for you because I do
mostly analog stuff. Sometimes it's the whole architecture though and
then one of my fuirst questions to the client is about parts sourcuiing
and longevity.

Who was it that said things should be done as simply as possible, but no
further. You are here because the analog approach is not so simple due
to the... well, analog effects.


Not quite. I can do it analog and with high precision. It's just that I
was hoping there'd be a solution in a can, in the form of a li'l class D
chip that happens to sport a good DC accuracy. That's why I am here, not
because an analog solution won't work. More because a good analog
solution consumes real estate but so does a processors with clock
crystal and all.

Usually digital is already in a system, it is just a question of how
much of it. That is another of the advantages. Once you open the
digital jar, you can put a lot in there.

--

Rick

 

On Sun, 01 Jun 2014 14:09:00 +0300, upsidedown@downunder.com wrote:

There are political risks as seen in Eastern Europe and Asia in the
last two decades. Also trade embargoes (such as CoCom) may limit into
what countries you can sell your products, if you are using components
originated from specific countries.

For 10+ year design, you really need to stay away from politically
unstable countries like (in alphabetic order) China, Russia or USA,

When selecting components or subsystems, these big countries could
block the availability of critical components due to political
reasons, either directly or indirectly.

While smaller countries might have local political problems, they do
not usually want loose the foreign trade, unless the society really
disintegrates (e.g. Yugoslavia).

 

[John Larkin]

On Sun, 01 Jun 2014 03:50:11 -0400, rickman <gnuarm@gmail.com> wrote:

On 5/31/2014 10:16 AM, Joerg wrote:

It was a cutting-edge FPGA in the mid 90's. With ASICs you can get much
more design security. The trick is to pick a foundry you trust and a
run-of-the-mills process that is used for tons of other products. By
going directly to the foundry you are cutting out one middleman (the
FPGA vendor) and thus reduce the overall risk.

I don't know what a "cutting edge" FPGA is other than a startup company.
Is that risky... yes of course.

You are speculating about the foundry vs. FPGA. All foundries are used
for "tons" of other products or they wouldn't stay in business. In
fact, the FPGA companies are the customers that define the foundry
business being the first real customers of the state of the art new fab
lines.

FPGAs are among the longest lived complex semiconductor devices out
there. You may find transitors, op amps and 8051s that live a lot
longer, but you won't find much LSI that lives longer than FPGAs.


An example for the longevity of semiconductor processes: We just started
something on a 4" wafer. Those date back almost to the days of
Methusaleh yet it's no problem. If a process is also used to make mil
stuff, chances are it'll survive both of us.

So? You can't find anything that competes with FPGAs in that technology.


And I might. Plus opamps/comparators.

Ok, but you are here because you are finding this approach to be a PITA
because of all the accuracy issues.


I pretty much only used the 8051 when I was allowed to decide. Clients
have used Atmels and also PICs that have lasted a long time but I don't
have production data for those products.

Yes, some MCUs live for a while, but not 20 years. I have never seen an
MCU other (than the 8051 and a few Mil/aerospace processors) that long.

One remarkable CPU is the Motorola/Freescale MC68332. It's just now going out of
production after at least 25 years.



--

John Larkin Highland Technology Inc
www.highlandtechnology.com jlarkin at highlandtechnology dot com

Precision electronic instrumentation

 

[Phil Hobbs]

On 6/1/2014 2:32 PM, John Larkin wrote:

On Sun, 01 Jun 2014 03:50:11 -0400, rickman <gnuarm@gmail.com> wrote:

On 5/31/2014 10:16 AM, Joerg wrote:

It was a cutting-edge FPGA in the mid 90's. With ASICs you can get much
more design security. The trick is to pick a foundry you trust and a
run-of-the-mills process that is used for tons of other products. By
going directly to the foundry you are cutting out one middleman (the
FPGA vendor) and thus reduce the overall risk.

I don't know what a "cutting edge" FPGA is other than a startup company.
Is that risky... yes of course.

You are speculating about the foundry vs. FPGA. All foundries are used
for "tons" of other products or they wouldn't stay in business. In
fact, the FPGA companies are the customers that define the foundry
business being the first real customers of the state of the art new fab
lines.

FPGAs are among the longest lived complex semiconductor devices out
there. You may find transitors, op amps and 8051s that live a lot
longer, but you won't find much LSI that lives longer than FPGAs.


An example for the longevity of semiconductor processes: We just started
something on a 4" wafer. Those date back almost to the days of
Methusaleh yet it's no problem. If a process is also used to make mil
stuff, chances are it'll survive both of us.

So? You can't find anything that competes with FPGAs in that technology.


And I might. Plus opamps/comparators.

Ok, but you are here because you are finding this approach to be a PITA
because of all the accuracy issues.


I pretty much only used the 8051 when I was allowed to decide. Clients
have used Atmels and also PICs that have lasted a long time but I don't
have production data for those products.

Yes, some MCUs live for a while, but not 20 years. I have never seen an
MCU other (than the 8051 and a few Mil/aerospace processors) that long.

One remarkable CPU is the Motorola/Freescale MC68332. It's just now going out of
production after at least 25 years.

Nice part--easy to use, powerful, lots of great peripherals (especially
for timing). I used it in an instrument design circa 1993.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

 

[Lasse Langwadt Christense]

Den mandag den 2. juni 2014 00.06.55 UTC+2 skrev Tim Wescott:

On Sat, 31 May 2014 04:41:06 -0400, rickman wrote:



On 5/30/2014 7:59 PM, Joerg wrote:



Stopband is a pretty gradual thing when you can't go past 3rd order on

the output filter. This application is sensitive to noise because the

load is super fast.



I guess I'm not really following all this. You have a control circuit

that is not easy to do with analog chips and yet you don't want to use a

digital control. From the bits and pieces I am seeing I believe a

classic ADC-DSP-DAC would suit this problem very well. Forget the silly

PWM thing and all the goofy filtering problems.



He hasn't said explicitly, but I'm 99.44% sure that he wants a PWM stage

for power efficiency reasons -- he's not just looking for a low-level

signal out, he's looking for some efficiently generated power at a finely

controlled voltage.



So a class D amplifier is the way to go.

I think quite a few of them are not really using classic pwm,
but more some sort of delta-sigma self-oscillating thing

If he needs to do the power drivers as well maybe something like
an IRS2092


-Lasse

 

[Tim Wescott]

On Sat, 31 May 2014 04:41:06 -0400, rickman wrote:

On 5/30/2014 7:59 PM, Joerg wrote:

Stopband is a pretty gradual thing when you can't go past 3rd order on
the output filter. This application is sensitive to noise because the
load is super fast.

I guess I'm not really following all this. You have a control circuit
that is not easy to do with analog chips and yet you don't want to use a
digital control. From the bits and pieces I am seeing I believe a
classic ADC-DSP-DAC would suit this problem very well. Forget the silly
PWM thing and all the goofy filtering problems.

He hasn't said explicitly, but I'm 99.44% sure that he wants a PWM stage
for power efficiency reasons -- he's not just looking for a low-level
signal out, he's looking for some efficiently generated power at a finely
controlled voltage.

So a class D amplifier is the way to go.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

 

On Sun, 1 Jun 2014 15:37:42 -0700 (PDT), Lasse Langwadt Christensen
<langwadt@fonz.dk> wrote:

Den mandag den 2. juni 2014 00.06.55 UTC+2 skrev Tim Wescott:
On Sat, 31 May 2014 04:41:06 -0400, rickman wrote:



On 5/30/2014 7:59 PM, Joerg wrote:



Stopband is a pretty gradual thing when you can't go past 3rd order on

the output filter. This application is sensitive to noise because the

load is super fast.



I guess I'm not really following all this. You have a control circuit

that is not easy to do with analog chips and yet you don't want to use a

digital control. From the bits and pieces I am seeing I believe a

classic ADC-DSP-DAC would suit this problem very well. Forget the silly

PWM thing and all the goofy filtering problems.



He hasn't said explicitly, but I'm 99.44% sure that he wants a PWM stage

for power efficiency reasons -- he's not just looking for a low-level

signal out, he's looking for some efficiently generated power at a finely

controlled voltage.



So a class D amplifier is the way to go.


I think quite a few of them are not really using classic pwm,
but more some sort of delta-sigma self-oscillating thing

The class-D (audio) amplifiers I'm aware of fall into two categories,
classic triangle/PWM ADC style and direct conversion (digital-PWM)
DSP. The first category far outnumbers the second (even with digital
input), at least for now. I don't know any that have the frequency
range he's looking for, though. All are intended to operate somewhere
around 300kHz to 500kHz, with a couple of new ones coming out at 2MHz.

If he needs to do the power drivers as well maybe something like
an IRS2092

 

[Maynard A. Philbrook Jr.]

In article <C6GdnafUpe5iPhbOnZ2dnUVZ_tidnZ2d@giganews.com>,
tim@seemywebsite.really says...

I guess I'm not really following all this. You have a control circuit
that is not easy to do with analog chips and yet you don't want to use a
digital control. From the bits and pieces I am seeing I believe a
classic ADC-DSP-DAC would suit this problem very well. Forget the silly
PWM thing and all the goofy filtering problems.

He hasn't said explicitly, but I'm 99.44% sure that he wants a PWM stage
for power efficiency reasons -- he's not just looking for a low-level
signal out, he's looking for some efficiently generated power at a finely
controlled voltage.

Years ago, and I think it may still be in use, I modified a designed on
a small low powered tape feeder. It had two small DC motors, one to pull
the spool at constant speed and the other for tension control.

The power supply and machine was small (compact) so the supply current
was limited. There were times where the tension would fight the speed
and a small oscillation would occur. After some debugging I found the
bus was becoming unstable when both motors required current at the same
time. This would thus make bus voltage unstable and cause control loop
problems.

Like I said, supply was limiting and hard to correct. The limited drive
circuit was a PWM for both, one had a Tach feed back for the first motor
and the other used the current as torque to regulate. I doubled the
clock rate for the PWM and put a Flip flop in there to operate the
drives out of phase from each other.

That corrected the problem. Something that makes PWM useful among other
things.

Jamie

 

[Joerg]

Lasse Langwadt Christensen wrote:

Den mandag den 2. juni 2014 00.06.55 UTC+2 skrev Tim Wescott:
On Sat, 31 May 2014 04:41:06 -0400, rickman wrote:



On 5/30/2014 7:59 PM, Joerg wrote:
Stopband is a pretty gradual thing when you can't go past 3rd order on
the output filter. This application is sensitive to noise because the
load is super fast.
I guess I'm not really following all this. You have a control circuit
that is not easy to do with analog chips and yet you don't want to use a
digital control. From the bits and pieces I am seeing I believe a
classic ADC-DSP-DAC would suit this problem very well. Forget the silly
PWM thing and all the goofy filtering problems.


He hasn't said explicitly, but I'm 99.44% sure that he wants a PWM stage

for power efficiency reasons -- he's not just looking for a low-level

signal out, he's looking for some efficiently generated power at a finely

controlled voltage.



So a class D amplifier is the way to go.


I think quite a few of them are not really using classic pwm,
but more some sort of delta-sigma self-oscillating thing

If he needs to do the power drivers as well maybe something like
an IRS2092

I was eyeing that one but it doesn't seem to have much DC stability and
no compensation for dead time drift. At its switching frequency of
800kHz that begins to matter.

No matter how I toss and tumble it, this begins to look like another
home-brew project.

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

krw@attt.bizz wrote:

On Sun, 1 Jun 2014 15:37:42 -0700 (PDT), Lasse Langwadt Christensen
langwadt@fonz.dk> wrote:

Den mandag den 2. juni 2014 00.06.55 UTC+2 skrev Tim Wescott:

[...]

He hasn't said explicitly, but I'm 99.44% sure that he wants a PWM stage

for power efficiency reasons -- he's not just looking for a low-level

signal out, he's looking for some efficiently generated power at a finely

controlled voltage.

Yup.

So a class D amplifier is the way to go.

I think quite a few of them are not really using classic pwm,
but more some sort of delta-sigma self-oscillating thing

The class-D (audio) amplifiers I'm aware of fall into two categories,
classic triangle/PWM ADC style and direct conversion (digital-PWM)
DSP. The first category far outnumbers the second (even with digital
input), at least for now. I don't know any that have the frequency
range he's looking for, though. All are intended to operate somewhere
around 300kHz to 500kHz, with a couple of new ones coming out at 2MHz.

That could even be acceptable if they had DC stability and dead-time
drift compensation (where they correct the OPWM accordingly).

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

Steve Wilson wrote:

Joerg <invalid@invalid.invalid> wrote:

Many analog/mixed things do not need FPGA. They compete differently.
Like a fast pulse-echo thing I just finished. It could be done in FPGA
but it's all analog, in this case for cost reasons. It is hard to beat
3-cent transistors.

I know you like to roll your own, but what value do you assign to system
design, board stuffing, checking for proper solder joints, on-board
diagnostics, troubleshooting, repair, documentation, procurement, incoming
inspection, etc.? It would seem a 3 cent transistor would end up costing
much more, but the question is how much more? And how do you calculate it?

You have to eyeball it because much also depends on the time the ICT
needs per board, affecting production throughput. I have a few boards
out there that are being in-circuit tested since day one and it isn't a
huge cost factor. It's in Asia but labor costs don't matter much with
ICT because it is largely automatic.

A huge factor is how well testability is designed into the board. To use
power PWM stages as an example, it can be good to add a couple
resistors and a cap so the ICT can gauge correct PWM operation all the
way to the end. Costs 1-2 cents extra but can speed up testing and allow
a cheaper ICT machine.


> The whole issue probably belongs in the BOM Minimization thread



--
Regards, Joerg

http://www.analogconsultants.com/

 

[Joerg]

rickman wrote:

On 6/1/2014 11:09 AM, Joerg wrote:
rickman wrote:
On 5/31/2014 10:16 AM, Joerg wrote:

It was a cutting-edge FPGA in the mid 90's. With ASICs you can get much
more design security. The trick is to pick a foundry you trust and a
run-of-the-mills process that is used for tons of other products. By
going directly to the foundry you are cutting out one middleman (the
FPGA vendor) and thus reduce the overall risk.

I don't know what a "cutting edge" FPGA is other than a startup company.
Is that risky... yes of course.


With cutting edge I meant the latest and gratest (newest) from a major
FPGA vendor. The usual, more gates, faster.

I don't know why you can't be specific about the vendor and part. ...

As I said it was a very long time ago, I was not the designer of this
board and I tend to purge my notes and brain cells once in a while, of
stuff that I no longer need. I remember that they did get enough notice
to redesign it but then, based on production quantities, decided ASIC.

... I
can't think of an example where one of the FPGA companies obsoleted a
part any time within a 10 year time frame. I have a recollection of
Xilinx dropping a package/part combo a bit earlier, but I'm not sure.

Dropping a package is all it takes. That blows your production straight
out of the water.

If you can go the ASIC route, then by all means do it. But FPGAs are
usually used where ASICs are cost prohibitive because of the low volume.

I remember when some of the Xilinx reps would post in c.a.fpga about how
the ASIC market would continue to shrink to nearly nothing because of
the high mask costs. ...

They were wrong. Just like the guy who told me 20 years ago that CD4000
logic was on the way out.

MPW and shuttle runs have actually made the entry fees into that world
lower. One of the ICs that we sucessfully designed together with an IC
house had less than $200k NRE and that included my time and a nice batch
of chips from a MPW run.

... But they were not comparing apples to oranges...
which they should have been. If you need today's technology in an FPGA
you can do the design in a 4 year old ASIC at least.

And then produce it almost forever. A run-of-the-mills BiCMOS process
these days will get the occasional additional module (such as
higher-voltage capable devices on the chip) but you don't have to use it.

[...]

An example for the longevity of semiconductor processes: We just
started
something on a 4" wafer. Those date back almost to the days of
Methusaleh yet it's no problem. If a process is also used to make mil
stuff, chances are it'll survive both of us.

So? You can't find anything that competes with FPGAs in that
technology.


Many analog/mixed things do not need FPGA. They compete differently.
Like a fast pulse-echo thing I just finished. It could be done in FPGA
but it's all analog, in this case for cost reasons. It is hard to beat
3-cent transistors.

Yes and you are here looking for a solution that you are finding
difficult in analog.

I do not find it difficult, I am just looking for something that has
this in an IC for less real estate use. That's all.

And I might. Plus opamps/comparators.

Ok, but you are here because you are finding this approach to be a PITA
because of all the accuracy issues.


I can get the accuracy alright. I am here because I couln't imagine
there not being a class-D modulator with DC accuracy. But it seems there
truly isn't one.

You seem to be contradicting yourself. I don't follow.

As I said, I can easily design this in discretes and opamps. If there
was a class-D modulator that has clean DC-handling I would not have to
and can save real estate on the board.

I pretty much only used the 8051 when I was allowed to decide. Clients
have used Atmels and also PICs that have lasted a long time but I don't
have production data for those products.

Yes, some MCUs live for a while, but not 20 years. I have never seen an
MCU other (than the 8051 and a few Mil/aerospace processors) that long.
Why? Because no one wants them. ...


Huh? I have many clients who want exactly that. They want a product they
ideally won't have to touch and can just keep selling. This is one of
the reasons why products such as pellet stoves and other reidential and
commercial gear use the 8051.

Yes, you have clients... but how many chips? Until the number reaches
millions per year no one cares and you are stuck with 8051s.

LOL, designing the 8051 into a pellet stove doesn't really mean a lot.
If the volumes were higher, they could save money going to the Asian 4
bitters. They are using the 8051 because they barely need an MCU. If
they need anything at all more complex they are stuck.

You can use a faster 8051 but yes, if you need more performance the ARM
is the way to go. Or even Atom. That will probably happen for me this year.

... They are old, slow and not cost
effective. If you seriously need a 20+ year production life without
redesign of any sort, then you will need to stick with 8051s and 4000
series CMOS. Is redesign really that big a problem that it can't be
done every 10 years or so?


In many cases, yes, it is undesirable.


I am pretty sure they will supply. They told me so and they've never
lied to me. Their prices are on the high side anyhow and I don't think
they would jack up the price on customers.

I have no idea what you base that on. ...


On trust. I have developed a trust level with that company over decades
and (in contrast to some others) they have never disappointed.

That means your needs have been compatible with their needs... so far.

I do a lot of very unorthodox designs with their chips. I prefer them
for two reasons:

a. LTSpice models.

b. Longevity in the supply chain.

... If you ask them to make a part
they have discontinued it will cost an arm and a leg. ...


Again, they call first and they won't discontinue if there is still a
product line that uses the chip. I assume they'd expect reasonable
quantities but that is generally the case with my designs. So it's not a
question of re-starting an old IC, it is one of not stopping it in the
first place.

I don't believe that for a minute. If you are the only customer left
they will keep the part in production? Really? Have they actually told
you that?

Yes, they have.

... Just give it a
thought for a bit. Running a fab is not cheap and making a run of parts
is not done for less than how many thousands of chips?


We are right now doing a run of about 10,000 or so, depending on the
yield. It's not that bad if you spread the cost over each resulting
device.

Most ASICs I have been involved in run between 100,000 and a million
parts per year which in the world of semiconductors is a drop in the
bucket. They are all in the low single-digit Dollar range. Order gets
called in, they make another round of wafers. Often they pre-make them
and just dice them up after the order comes in. So you have to give them
a reasonable forecast. It's almost like running a catering business.

Now you are mixing apples and oranges again. ASICs are typically not
full custom. Your custom ASIC is shared across many customers with a
small number of custom layers. The ADI chips we were discussing are
full custom parts and running a batch is a big deal. I am sure if you
were the last customer for a part you would have to go pound sand.

You are probably thinking of ASIC in their old definition from the 80's
or so where the digital ones were in essence like gate array. Nowadays
you can have tons of analog standard funtions and they get scaled to
your needs. You can also design your own blocks. BTDT, many times. It's
full custom and some of the chip design places even have ASIC in their
names, like this one whioch was recently bought by Microsemi:

http://investing.businessweek.com/research/stocks/private/snapshot.asp?privcapId=24432205

All the ones I ever dealt with will design a fully tailored IC for you
and, unless this is explicitly agreed upon to save NRE or whatever, it
will not be shared with any other customer of theirs.

[...]

This is one reason why I try to keep things simple. Many jobs that are
done with a DSP can be done without. But not all of them, of course. Of
course, for me that's easier to say than probably for you because I do
mostly analog stuff. Sometimes it's the whole architecture though and
then one of my fuirst questions to the client is about parts sourcuiing
and longevity.

Who was it that said things should be done as simply as possible, but no
further. You are here because the analog approach is not so simple due
to the... well, analog effects.


Not quite. I can do it analog and with high precision. It's just that I
was hoping there'd be a solution in a can, in the form of a li'l class D
chip that happens to sport a good DC accuracy. That's why I am here, not
because an analog solution won't work. More because a good analog
solution consumes real estate but so does a processors with clock
crystal and all.

Usually digital is already in a system, it is just a question of how
much of it. That is another of the advantages. Once you open the
digital jar, you can put a lot in there.

Well, no digital jar on this one unless I put it there ;-)

--
Regards, Joerg

http://www.analogconsultants.com/