"Mike Engelhardt has parted ways with Analog Devices"

On Sunday, March 8, 2020 at 3:24:35 PM UTC+11, jla...@highlandsniptechnology.com wrote:
> On Sun, 8 Mar 2020 04:01:13 -0000 (UTC), Przemek Klosowski
> <przemek@tux.dot.org> wrote:
> 
> >On Wed, 04 Mar 2020 19:19:12 -0800, Rick C wrote:
> >
> >> On Wednesday, March 4, 2020 at 9:55:22 PM UTC-5, Simon S Aysdie wrote:
> >>> "This does not bode well for the future of LTspice now that Mike
> >>> Engelhardt has parted ways with Analog Devices."---analog spiceman
> >>> 
> >>> 
> >>> https://groups.io/g/LTspice/topic/cloud_over_ltspice/71467233
> >> 
> >> Time will tell.  I don't know how ADI views LTspice.  It's a free tool
> >> that likely provides only a minimum return.  Maybe they will find
> >> someone else to take it over.  Or maybe they will allow it to continue
> >> without much further effort.
> >> 
> >> Probably the only real work it requires is to support models for new
> >> parts.  I'd be willing to bet that is already handled by the groups who
> >> produce the various parts.
> >
> >Mike's 2018 interview says that his main job at AD is creating AD part 
> >models in LTSpice
> >
> >https://www.youtube.com/watch?v=x6TrbD7-IwU
> 
> "Intuition is the most important part of engineering."
> 
> That's great.

It's actually insane, but if intuition is all you've got, you might agree.

-- 
Bill Sloman, Sydney

On Sunday, March 8, 2020 at 3:50:29 PM UTC+11, Rick C wrote:
> On Saturday, March 7, 2020 at 11:22:27 PM UTC-5, Bill Sloman wrote:
> > On Sunday, March 8, 2020 at 2:19:06 PM UTC+11, Rick C wrote:
> > > On Saturday, March 7, 2020 at 8:13:49 PM UTC-5, Bill Sloman wrote:
> > > > On Sunday, March 8, 2020 at 10:22:33 AM UTC+11, Uwe Bonnes wrote:
> > > > > Bill Sloman <bill.sloman@ieee.org> wrote:
> > > > > ...
> > > > > > 
> > > > > > Not altogether surprising. The process Analog Devices uses to
> > > > > manufacture the AD797 is remarkably fancy. 
> > > > > > 
> > > > > Can you shed some light?
> > > > 
> > > > Not a lot. I can recall talk of ion-implanted PNP transistors that were really fast, high gain parts, and some comment to the effect that it was a 20-stage process, where regular processes were about 12 stages, but it was all from some Analog Devices presentation, and I haven't been to one for a decade or so.
> > > > 
> > > > Googling did reveal that Analog Devices still make their high-end analog parts and MEMS parts in-house, rather than relying on foundries, but nothing all that specific.
> > > 
> > > I would expect that given the equipment doesn't get out of date as fast as the state of the art stuff used for making advanced digital chips.
> > 
> > They might have less of a drive to get the highest possible resolution, and anyway ion implantation can offer you that rather more cheaply than optical techniques - and has been able to do that for ages. It's as slow as a wet week, so you can only have the high resolution at a few critical spots on the device, which is probably all you need for something like the AD797
> 
> You haven't said anything about the technology required to make the parts.

I did mention ion-implantation.

> That's my point.  I think continuing to advance analog design doesn't take the same level of investment as it does for digital designs.  
> 
> > Analog parts do tend to be less modular than digital parts.
> > 
> > You probably need to think about the state of the relevant art, rather than thinking that making big fast digital chips is the only art that matters.
> 
> How about sticking with what I said rather than what you imagine I said.  I never said anything about the worth of design art.  I was talking solely about the cost of production.  Most analog designs have fundamental limitations in how small they can be due to power handling.

Some do. Cambridge Intruments sold an electron beam microfabricator to Thompson in the early 1980s.

The acceptance test was to direct write the very narrow gates on three wafers of very high frequency FETs.

Someone looked at the price Thompson CSF was charging for those FETs and those three wafers would have paid the million dollar price of the machine.

>  It doesn't really matter so much how many transistors you stuff into the innards of a regulator because the area of the chip will be dominated by the power handling components.  So not as much incentive to continue to miniaturize.

  
Regulators aren't what we were talking about. If you need fine-line structures to get local speed, your argument falls flat.

> > >  So they can continue pounding out the same parts using fully depreciated assets.  
> > 
> > That's probably not the way they see it.
> 
> You mean they buy new fab lines just so they can start the depreciation cycle over again?

Incremental development doesn't necessarily involve buying a new fab line.

>  Even then, the costs of starting a typical new analog design is so much less than a digital design that they can afford to have many devices with much lower production volumes.  

The cost of working out how to get better performance out of a nominally standard analog design can be appreciable - you may be talking about less numerous elements, but you have to be much pickier about how each elements behaves.
 
> You really haven't addressed the meat of anything I've mentioned.

Not in terms you seem to be willing to think about.  
 
> Analog designs are cheaper to start and cheaper to produce.

They can be. The step up from the LT1028 to the AD797 (which is actually a simpler circuit) may not have been all that cheap.

> So it is more practical to have lots of designs that aren't super high volume which is simply not the case for digital designs that are anywhere at all near state of the art. 

The state of the art for digital designs is not inter-changable with the state of the art for analog designs. You've got different constraints and different targets. A fab line optimised to produce high performance analog isn't going to look much like a fab line optimised to produce any of the various flavours of high performance digital parts.

And of course Analog Devices has a separate fab line designed to produce MEMS (micro-electro-mechanical) parts.

-- 
Bill Sloman, Sydney

-- 
Bill Sloman, Sydney

On Sunday, March 8, 2020 at 1:06:08 AM UTC-5, Bill Sloman wrote:
> On Sunday, March 8, 2020 at 3:50:29 PM UTC+11, Rick C wrote:
> > On Saturday, March 7, 2020 at 11:22:27 PM UTC-5, Bill Sloman wrote:
> > > On Sunday, March 8, 2020 at 2:19:06 PM UTC+11, Rick C wrote:
> > > > On Saturday, March 7, 2020 at 8:13:49 PM UTC-5, Bill Sloman wrote:
> > > > > On Sunday, March 8, 2020 at 10:22:33 AM UTC+11, Uwe Bonnes wrote:
> > > > > > Bill Sloman <bill.sloman@ieee.org> wrote:
> > > > > > ...
> > > > > > > 
> > > > > > > Not altogether surprising. The process Analog Devices uses to
> > > > > > manufacture the AD797 is remarkably fancy. 
> > > > > > > 
> > > > > > Can you shed some light?
> > > > > 
> > > > > Not a lot. I can recall talk of ion-implanted PNP transistors that were really fast, high gain parts, and some comment to the effect that it was a 20-stage process, where regular processes were about 12 stages, but it was all from some Analog Devices presentation, and I haven't been to one for a decade or so.
> > > > > 
> > > > > Googling did reveal that Analog Devices still make their high-end analog parts and MEMS parts in-house, rather than relying on foundries, but nothing all that specific.
> > > > 
> > > > I would expect that given the equipment doesn't get out of date as fast as the state of the art stuff used for making advanced digital chips.
> > > 
> > > They might have less of a drive to get the highest possible resolution, and anyway ion implantation can offer you that rather more cheaply than optical techniques - and has been able to do that for ages. It's as slow as a wet week, so you can only have the high resolution at a few critical spots on the device, which is probably all you need for something like the AD797
> > 
> > You haven't said anything about the technology required to make the parts.
> 
> I did mention ion-implantation.

Yes, you mentioned it again, but neither time was relevant to the issue. 


> > That's my point.  I think continuing to advance analog design doesn't take the same level of investment as it does for digital designs.  
> > 
> > > Analog parts do tend to be less modular than digital parts.
> > > 
> > > You probably need to think about the state of the relevant art, rather than thinking that making big fast digital chips is the only art that matters.
> > 
> > How about sticking with what I said rather than what you imagine I said.  I never said anything about the worth of design art.  I was talking solely about the cost of production.  Most analog designs have fundamental limitations in how small they can be due to power handling.
> 
> Some do. Cambridge Intruments sold an electron beam microfabricator to Thompson in the early 1980s.
> 
> The acceptance test was to direct write the very narrow gates on three wafers of very high frequency FETs.
> 
> Someone looked at the price Thompson CSF was charging for those FETs and those three wafers would have paid the million dollar price of the machine.
> 
> >  It doesn't really matter so much how many transistors you stuff into the innards of a regulator because the area of the chip will be dominated by the power handling components.  So not as much incentive to continue to miniaturize.
> 
>   
> Regulators aren't what we were talking about. If you need fine-line structures to get local speed, your argument falls flat.

Not really.  What analog production devices are you suggesting have that sort of fine line requirement that digital devices do today? 


> > > >  So they can continue pounding out the same parts using fully depreciated assets.  
> > > 
> > > That's probably not the way they see it.
> > 
> > You mean they buy new fab lines just so they can start the depreciation cycle over again?
> 
> Incremental development doesn't necessarily involve buying a new fab line.

Exactly!  In the digital domain they have no choice but to build entirely new fab lines every year or two.  With the technology required to build analog devices this is not so much the case and nothing remotely like the expense of the digital fab lines is needed. 


> >  Even then, the costs of starting a typical new analog design is so much less than a digital design that they can afford to have many devices with much lower production volumes.  
> 
> The cost of working out how to get better performance out of a nominally standard analog design can be appreciable - you may be talking about less numerous elements, but you have to be much pickier about how each elements behaves.

Which has very little to do with the issue at hand.  The bottom line is designing and fabrication costs of analog are much, much lower than today's digital chips. 


> > You really haven't addressed the meat of anything I've mentioned.
> 
> Not in terms you seem to be willing to think about.  

Not in any meaningful terms.  You are just bobbing and weaving without saying anything meaningful. 


> > Analog designs are cheaper to start and cheaper to produce.
> 
> They can be. The step up from the LT1028 to the AD797 (which is actually a simpler circuit) may not have been all that cheap.

Yes, I'm sure it was not "all that cheap".  But compared to the digital designs today it was a pittance which is my point.  Maybe the AD797 is a chip that can only be profitable at higher volumes, I don't know.  But most of the analog chips ADI and LT make are not high volume chips but are still profitable because of the lower costs than with many digital devices. 


> > So it is more practical to have lots of designs that aren't super high volume which is simply not the case for digital designs that are anywhere at all near state of the art. 
> 
> The state of the art for digital designs is not inter-changable with the state of the art for analog designs. You've got different constraints and different targets. A fab line optimised to produce high performance analog isn't going to look much like a fab line optimised to produce any of the various flavours of high performance digital parts.

I never said they were equivalent.  I said the costs of maintaining state of the art digital processing requires huge volumes which aren't required for most analog designs. 


> And of course Analog Devices has a separate fab line designed to produce MEMS (micro-electro-mechanical) parts.

More red herrings. 

Ok, I think we have covered this completely.  Continue on your own if you wish.  You aren't adding anything to the conversation at this point. 

-- 

  Rick C.

  +- Get 1,000 miles of free Supercharging
  +- Tesla referral code - https://ts.la/richard11209

On Sunday, March 8, 2020 at 6:04:31 PM UTC+11, Rick C wrote:
> On Sunday, March 8, 2020 at 1:06:08 AM UTC-5, Bill Sloman wrote:
> > On Sunday, March 8, 2020 at 3:50:29 PM UTC+11, Rick C wrote:
> > > On Saturday, March 7, 2020 at 11:22:27 PM UTC-5, Bill Sloman wrote:
> > > > On Sunday, March 8, 2020 at 2:19:06 PM UTC+11, Rick C wrote:
> > > > > On Saturday, March 7, 2020 at 8:13:49 PM UTC-5, Bill Sloman wrote:
> > > > > > On Sunday, March 8, 2020 at 10:22:33 AM UTC+11, Uwe Bonnes wrote:
> > > > > > > Bill Sloman <bill.sloman@ieee.org> wrote:
> > > > > > > ...
> > > > > > > > 
> > > > > > > > Not altogether surprising. The process Analog Devices uses to
> > > > > > > manufacture the AD797 is remarkably fancy. 
> > > > > > > > 
> > > > > > > Can you shed some light?
> > > > > > 
> > > > > > Not a lot. I can recall talk of ion-implanted PNP transistors that were really fast, high gain parts, and some comment to the effect that it was a 20-stage process, where regular processes were about 12 stages, but it was all from some Analog Devices presentation, and I haven't been to one for a decade or so.
> > > > > > 
> > > > > > Googling did reveal that Analog Devices still make their high-end analog parts and MEMS parts in-house, rather than relying on foundries, but nothing all that specific.
> > > > > 
> > > > > I would expect that given the equipment doesn't get out of date as fast as the state of the art stuff used for making advanced digital chips.
> > > > 
> > > > They might have less of a drive to get the highest possible resolution, and anyway ion implantation can offer you that rather more cheaply than optical techniques - and has been able to do that for ages. It's as slow as a wet week, so you can only have the high resolution at a few critical spots on the device, which is probably all you need for something like the AD797
> > > 
> > > You haven't said anything about the technology required to make the parts.
> > 
> > I did mention ion-implantation.
> 
> Yes, you mentioned it again, but neither time was relevant to the issue. 

Which is to say that you couldn't see the relevance.

The AD797 does seem to use high gain fast PNP transistors, and ion implantation does seem to be the way that Analog Devices gets them. It probably wouldn't be the kind of solution that would work for a digital array of millions of identical sets of transistors, but it's fine for this application. 

<snip>

-- 
Bill Sloman, Sydney

On 2020-03-07 22:19, Rick C wrote:
> On Saturday, March 7, 2020 at 8:13:49 PM UTC-5, Bill Sloman wrote:
>> On Sunday, March 8, 2020 at 10:22:33 AM UTC+11, Uwe Bonnes wrote:
>>> Bill Sloman <bill.sloman@ieee.org> wrote: ...
>>>> 
>>>> Not altogether surprising. The process Analog Devices uses to
>>> manufacture the AD797 is remarkably fancy.
>>>> 
>>> Can you shed some light?
>> 
>> Not a lot. I can recall talk of ion-implanted PNP transistors that
>> were really fast, high gain parts, and some comment to the effect
>> that it was a 20-stage process, where regular processes were about
>> 12 stages, but it was all from some Analog Devices presentation,
>> and I haven't been to one for a decade or so.
>> 
>> Googling did reveal that Analog Devices still make their high-end
>> analog parts and MEMS parts in-house, rather than relying on
>> foundries, but nothing all tht specific.
> 
> I would expect that given the equipment doesn't get out of date as
> fast as the state of the art stuff used for making advanced digital
> chips.  So they can continue pounding out the same parts using fully
> depreciated assets.

Analogue CMOS stopped getting better at about the 180-nm node, and was 
certainly going into the tank by 65 nm.  At that resolution, you can use 
single-exposure DUV or excimer litho and regular chrome masks.  That 
saves a lot of money.

Cheers

Phil Hobbs


-- 
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

Przemek Klosowski.... thank you for pointing to that interview on YouTube. I really liked the guys view on what the ltspice program is for which is so that the engineer can get better intuition as to how his circuits work. It is a piece of verification but should not be used is the sole purpose of verifying. I guess I like it because it resonated with my viewpoints about use of such a program

On 3/9/2020 6:14 AM, bulegoge@columbus.rr.com wrote:
> Przemek Klosowski.... thank you for pointing to that interview on YouTube. I really liked the guys view on what the ltspice program is for which is so that the engineer can get better intuition as to how his circuits work. It is a piece of verification but should not be used is the sole purpose of verifying. I guess I like it because it resonated with my viewpoints about use of such a program
> 

+1

On Mon, 9 Mar 2020 04:14:07 -0700 (PDT), bulegoge@columbus.rr.com
wrote:

>Przemek Klosowski.... thank you for pointing to that interview on YouTube. I really liked the guys view on what the ltspice program is for which is so that the engineer can get better intuition as to how his circuits work. It is a piece of verification but should not be used is the sole purpose of verifying. I guess I like it because it resonated with my viewpoints about use of such a program

Being an IC guy, he did miss that a little as regards PC boards. I
often use LT Spice as the only process before I go to a PC board, but
then PC boards can be modified a lot easier than linear ICs, and parts
are mostly temperature stable as purchased.

And I design using LT Spice. Once one has some intuition, one can just
throw parts around in the sim and see what happens. That works
surprisingly well. I have several circuits in production that I don't
really understand. 

I do much less math than I used to do. I guess rough values and tweak
in LT Spice. Voltage dividers, filters, oscillators this week. So it's
a calculator, too.

It also draws presentable diagrams to include in emails and manuals.
It's a drawing program.

I've had his same thought before: Romans built waterworks, people
built bridges and cathedrals and cannons and sailing ships, before
Newton invented calculus. Most science explained what people had
already built.







-- 

John Larkin         Highland Technology, Inc

The cork popped merrily, and Lord Peter rose to his feet. 
"Bunter", he said, "I give you a toast. The triumph of Instinct over Reason"

On 3/9/2020 9:55 AM, jlarkin@highlandsniptechnology.com wrote:
> On Mon, 9 Mar 2020 04:14:07 -0700 (PDT), bulegoge@columbus.rr.com
> wrote:
> 
>> Przemek Klosowski.... thank you for pointing to that interview on YouTube. I really liked the guys view on what the ltspice program is for which is so that the engineer can get better intuition as to how his circuits work. It is a piece of verification but should not be used is the sole purpose of verifying. I guess I like it because it resonated with my viewpoints about use of such a program
> 
> Being an IC guy, he did miss that a little as regards PC boards. I
> often use LT Spice as the only process before I go to a PC board, but
> then PC boards can be modified a lot easier than linear ICs, and parts
> are mostly temperature stable as purchased.
> 
> And I design using LT Spice. Once one has some intuition, one can just
> throw parts around in the sim and see what happens. That works
> surprisingly well. I have several circuits in production that I don't
> really understand.

I don't throw parts around. I design first and then see if my design in 
LTSpice shows something I overlooked. If it doesn't result in what I 
expected, I try to learn why and correct it.

> I do much less math than I used to do. I guess rough values and tweak
> in LT Spice. Voltage dividers, filters, oscillators this week. So it's
> a calculator, too.

Indeed. I do the same. Sometimes it is faster than using a calculator.

> It also draws presentable diagrams to include in emails and manuals.
> It's a drawing program.

Agreed. I miss Mike.

On Mon, 9 Mar 2020 10:17:56 -0500, John S <Sophi.2@invalid.org> wrote:

>On 3/9/2020 9:55 AM, jlarkin@highlandsniptechnology.com wrote:
>> On Mon, 9 Mar 2020 04:14:07 -0700 (PDT), bulegoge@columbus.rr.com
>> wrote:
>> 
>>> Przemek Klosowski.... thank you for pointing to that interview on YouTube. I really liked the guys view on what the ltspice program is for which is so that the engineer can get better intuition as to how his circuits work. It is a piece of verification but should not be used is the sole purpose of verifying. I guess I like it because it resonated with my viewpoints about use of such a program
>> 
>> Being an IC guy, he did miss that a little as regards PC boards. I
>> often use LT Spice as the only process before I go to a PC board, but
>> then PC boards can be modified a lot easier than linear ICs, and parts
>> are mostly temperature stable as purchased.
>> 
>> And I design using LT Spice. Once one has some intuition, one can just
>> throw parts around in the sim and see what happens. That works
>> surprisingly well. I have several circuits in production that I don't
>> really understand.
>
>I don't throw parts around. I design first and then see if my design in 
>LTSpice shows something I overlooked. If it doesn't result in what I 
>expected, I try to learn why and correct it.

My 1200 volt 5 MHz Pockels Cell pulser resulted from throwing parts
around. My current LC oscillator is almost as bad. I didn't understand
it until the simulation, and a breadboard, were working. Intuition
guides throwing. You can't design circuits by truly random fiddling
because the solution space is too big.

"Something I overlooked" could be an entirely new concept.

>
>> I do much less math than I used to do. I guess rough values and tweak
>> in LT Spice. Voltage dividers, filters, oscillators this week. So it's
>> a calculator, too.
>
>Indeed. I do the same. Sometimes it is faster than using a calculator.

Something like a 5-resistor voltage divider or opamp circuit, made
from parts in stock, is way too hard to design on paper.



-- 

John Larkin         Highland Technology, Inc

The cork popped merrily, and Lord Peter rose to his feet. 
"Bunter", he said, "I give you a toast. The triumph of Instinct over Reason"