Talking with Steve Sanghi, CEO of Microchip for 31 years

[EEVblog]

Catching up with Steve Sanghi who was Microchip CEO for 31 years, now executive chariman.
Check out his new book, Up and to the Right. https://amzn.to/3M3DXFM
How Microchip was built into one of the biggest and most pofitable chip companies in the world.

This is copied from The Amp Hour podcast, as it shuld be of great interest to EEVblog viewers.
https://theamphour.com/

In this episode we discussed:

The early microcontrollers, ROM, EPROM, EEPROM, FLASH.
Semiconductor Fab's and locations.
How the US Government CHIPS act is a golden handcuffs trap.
US/China rhetoric.
What really happened durign the covid supply chain crisis.
How 90% of medical devices use Microchip parts.
Lead times.
Respect for the chip manufacturers and supply chains.
Just-In-Time becomes Just-In-Case.
How the Toyota production system works and how it had to change.
The impacts of a potential war in Taiwan.
How does it make sense to make thousands of part variants?
The secret to making part varients.
Why Microchip chose MIPS vs ARM.
Atmel didn't make any money!
Pricing discipline.
How buying Atmel almost didn't happen, and how Dialog Semiconductor goofed it.
Draconian NDA terms.
Atmel was bloated.
RISC-V plans.
Customer Driven Obsolescence.
Foundry vs In-House limitations.
Open Source FPGA tools?
Third party tool support.
Why not offer free optimised compilers?
It's all about black swan events, down cycles, pointy haired bosses, and how All your Cost Bases Beyong To Me.
Hobbyist vs professionals.
How Microchip is the largest aerospace chip maker in the world. Nothing leaves earth without a Microchip part in it.
Radiation hardended parts.
Market value profitability.
Will Microchip ever get acquired.
Strained US vs China relations.

[JPortici]

Looking forward on this one! the topic list is particularly juicy (did not know about aerospace, and did it happen before or after atmel+microsemi acquisitions? as it don't recall microchip ever having radhard parts before)

[EEVblog]

Looking forward on this one! the topic list is particularly juicy (did not know about aerospace, and did it happen before or after atmel+microsemi acquisitions? as it don't recall microchip ever having radhard parts before)

Not sure about that timeline.

[JPortici]

Atmel was acquired in 2016, microsemi in 2018
The first radhard parts i've heard from microchip were rebadges from those two company, that's what i'm saying.
Anyway, now i'm at the office so i can hear the podcast

[EEVblog]

Atmel was acquired in 2016, microsemi in 2018
The first radhard parts i've heard from microchip were rebadges from those two company, that's what i'm saying.

Seems not, been doing it since 1985
https://ww1.microchip.com/downloads/en/DeviceDoc/00002200C.pdf

ASIC System Solution with Mixed Signal Capabilities
Fully designed, assembled, tested and qualified in Europe since 1985, Microchip's offering of Rad-Hard Digital Application-Specific
Integrated Circuits (ASICs) is one of the most attractive and competitive on the market for the most critical applications.
To meet next-generation satellites and launchers needs, our ATMX150RHA improves integration with adaptable technology: a
mixed-signal strategy with Logic, Analog, 5V compatibility, embedded Non-Volatile Memory (NVM) option, high-voltage option and
qualified IPs.

[2N3055]

Atmel was acquired in 2016, microsemi in 2018
The first radhard parts i've heard from microchip were rebadges from those two company, that's what i'm saying.

Seems not, been doing it since 1985
https://ww1.microchip.com/downloads/en/DeviceDoc/00002200C.pdf

ASIC System Solution with Mixed Signal Capabilities
Fully designed, assembled, tested and qualified in Europe since 1985, Microchip's offering of Rad-Hard Digital Application-Specific
Integrated Circuits (ASICs) is one of the most attractive and competitive on the market for the most critical applications.
To meet next-generation satellites and launchers needs, our ATMX150RHA improves integration with adaptable technology: a
mixed-signal strategy with Logic, Analog, 5V compatibility, embedded Non-Volatile Memory (NVM) option, high-voltage option and
qualified IPs.

Those are Atmel...

[woofy]

Wow! Fascinating podcast, thanks Dave.

On RISC-V, all I can say is I would use them as I'm pretty agnostic when it comes to the CPU core, it just has to do the job. I program in C so the core is mostly invisible anyway.

[JPortici]

Atmel was acquired in 2016, microsemi in 2018
The first radhard parts i've heard from microchip were rebadges from those two company, that's what i'm saying.

Seems not, been doing it since 1985
https://ww1.microchip.com/downloads/en/DeviceDoc/00002200C.pdf

ASIC System Solution with Mixed Signal Capabilities
Fully designed, assembled, tested and qualified in Europe since 1985, Microchip's offering of Rad-Hard Digital Application-Specific
Integrated Circuits (ASICs) is one of the most attractive and competitive on the market for the most critical applications.
To meet next-generation satellites and launchers needs, our ATMX150RHA improves integration with adaptable technology: a
mixed-signal strategy with Logic, Analog, 5V compatibility, embedded Non-Volatile Memory (NVM) option, high-voltage option and
qualified IPs.

"With over 30 years of experience in the space industry from Atmel heritage" etc

[nctnico]

Yep. The actual Microchip parts are pretty crappy. I keep designing these out due to various reliability and functional issues.

Over the years Microchip has acquired some higher quality product ranges which are now branded Microchip. It is like Harbour Freight buying Festool and branding all Festool products as Harbour Freight...

[KE5FX]

Yep. The actual Microchip parts are pretty crappy. I keep designing these out due to various reliability and functional issues.

Over the years Microchip has acquired some higher quality product ranges which are now branded Microchip. It is like Harbour Freight buying Festool and branding all Festool products as Harbour Freight...

What are some examples of parts you've had problems with?

[SiliconWizard]

Yep. The actual Microchip parts are pretty crappy.

Oh really.

[PCB.Wiz]

On RISC-V, all I can say is I would use them as I'm pretty agnostic when it comes to the CPU core, it just has to do the job. I program in C so the core is mostly invisible anyway.

I know other vendors doing the same as Microchip - they use RISC V cores in the 'buried' stuff, typically the Bluetooth core, whilst the customer facing MCU is an ARM core.
I guess that has two benefits

• They can tune the RISC V they need for that task
• They avoid paying more mony

[globoy]

That was very interesting!  Especially for those of us who cut our teeth with PICs.  Makes me want to figure out some project for a 16C57, just for old time's sake...  Back in the day where you could memorize the entire instruction set and peripheral control registers.

It wasn't really discussed but I always thought one reason the PIC was so successful - and widely designed in - back in the day was because of cheap programmers, free samples and a free [assembly language] development environment.  At least for me getting my start in embedded that was a low barrier for entry compared with the other vendors.  Because of the simple instruction set the assembly programming didn't seem so onerous back then.

[schmitt trigger]

This topic is specially close to me as I was working for another division of General Instruments in the 1980s and witnessed first hand the fate of the Microelectronics division, and the struggles it had when it was initially spun-off.
Since we had several legacy products which used Microelectronics products, we had to deal with the spin-off glitches.

I agree that Mr. Sanghi was the right man at the right place and time, which transitioned an ailing company into the powerhouse that Microchip eventually became.

[woofy]

I still have some of the old PIC's from Yester-Year.

[AndyC_772]

Re: charging for optimised compilers, it's frightening to hear just how far out of touch he is from the engineers designing in his parts.

Anyone - large or small business - designing with a microcontroller needs to not only complete the design once, but also provide support for it over its lifetime. It's no good renting a compiler to complete a project, only to find a couple of months later an update is required, and then another one in 6 months, and another the following year, shortly followed by a port to a new device and so on.

Software maintenance is a thing. Bugs happen. Feature creep happens. Sometimes, parts disappear off the shelves for months at a time and have to be substituted. (Sound familiar?)

And the excuse for charging for them? Something something general managers blah blah blah. Seriously, WTF?

Sorry Steve, you're living in world in which ARM exists and GCC is free. The only thing keeping me using PICs these days is the 150C temperature rating.

[woofy]

Optimisation levels 0, 1 & 2 are included in the free compiler, only 3 & s require the paid for version.

[mariush]

I listened to the show this morning at work and my first impression was that... I have expected more from the discussion. Too little actual information, very brief on lots of questions, kinda disappointed really.

I mean seriously, when you have a person as big as this one on the show, couldn't you have made an exception and do it 2-3 hours interview? Though I doubt he would have had answers...

Almost like he only stayed for one hour to plug his book at the end... I mean kinda the nerve of the guy, he's probably a millionaire yet he still has to shill his book....  f#*cking 45$ for 300 pages, and self-published book, so it's not like a publisher get 50% or whatever of it.
 
The "we have to charge for compiler because otherwise we may sell the division when it doesn't money" is a ridiculous justification.

[zilp]

Yeah, his views regarding paid vendor dev tools seemed wildly out of touch indeed.

But not only because of bugfixes and stuff, but even just the insanity of managing licences, and possibly integrating that into an automated build system, possibly even some sort of continous integration setup, ... when you can just install GCC (or LLVM) without any restrictions whatsoever and have it available for automated release processes whenever you need it.

Plus, GCC being Free Software, all your compiler-specific skills are not locked into the manufacturer's ecosystem, and it only needs an additional code generator backend to make them even transferable to new, competing, architectures, and existing code is also easier to port to new targets this way as the compiler's language features don't depend on the target.

So, if you ask me, asking for money for vendor dev tools in this environment is much more likely to kill them than giving them away for free, as there are significant disincentives to use them already anyway, and I would think minor optimization advantages would rarely justify all these disadvantages, especially so given that a bunch more computing power or memory rarely costs significantly more anymore, so unless you are dealing with a very cost sensitive and/or large volume product, who would deal with that hassle instead of just selecting a slightly larger chip?

[KE5FX]

To play devil's advocate, if I'm building 10,000,000 of something and a compiler that's 1% better than GCC lets me ship with a PIC that costs 10 cents less, that compiler license is worth up to $1 million.

Of course, there are likely to be any number of ways to shave 1% from the code size without paying a six-figure sum for a compiler.  I guess he's assuming that a lot of customers don't have the time or talent needed to find another way.

[EEVblog]

Yeah, his views regarding paid vendor dev tools seemed wildly out of touch indeed.

They might be "out of touch" when it comes to small time users, but name a microcontroller maker more consistently profitable than Microchip. Also name another microcontroller manufacturer that still makes and supports ancient parts as ell as Microchip. Perhaps beig "out of touch" is the price paid for profitability and longevity?

[EEVblog]

I listened to the show this morning at work and my first impression was that... I have expected more from the discussion. Too little actual information, very brief on lots of questions, kinda disappointed really.
I mean seriously, when you have a person as big as this one on the show, couldn't you have made an exception and do it 2-3 hours interview? Though I doubt he would have had answers...

It was intially going to be a private phone call, like the last time we talked. Then he said he could swing half an hour and do the recording for the Amp Hour. Then I pushed for the full hour and eventually got an hour slot in his calendar. He's a busy guy.
I think there is a ton of useful and interesting stuff in the hour we got. Just look at the list of subjects we covered for goodness sake! Pretty darn impressive for 1 hour even if I do say myself.
And this is why the audio sucks. I didn't have the time to get things set up right on his end because it would have eaten into the available time. And then when the audio recording failed 3/4 the way through I had to push on and hope I could get the recovery file from Zencaster. Luckily I did.
Be thankful I got the hour worth of material.

Almost like he only stayed for one hour to plug his book at the end... I mean kinda the nerve of the guy, he's probably a millionaire yet he still has to shill his book....  f#*cking 45$ for 300 pages, and self-published book, so it's not like a publisher get 50% or whatever of it.

So don't buy it, or buy the Kindle edition.

The "we have to charge for compiler because otherwise we may sell the division when it doesn't money" is a ridiculous justification.

It actually makes sense.

[EEVblog]

To play devil's advocate, if I'm building 10,000,000 of something and a compiler that's 1% better than GCC lets me ship with a PIC that costs 10 cents less, that compiler license is worth up to $1 million.
Of course, there are likely to be any number of ways to shave 1% from the code size without paying a six-figure sum for a compiler.  I guess he's assuming that a lot of customers don't have the time or talent needed to find another way.

It's $44/month.

[zilp]

They might be "out of touch" when it comes to small time users, but name a microcontroller maker more consistently profitable than Microchip. Also name another microcontroller manufacturer that still makes and supports ancient parts as ell as Microchip. Perhaps beig "out of touch" is the price paid for profitability and longevity?

Uh ... or maybe the long-term availability is part of what makes them successful, and they'd be more so if they had a different approach to dev tool licensing?

And also, it's not like you couldn't target any of their chips with free compilers ... and, after all, even their XC* compilers are GCC under the hood, just with questionable handling of the licensing, so arguably, they are free compilers, just with some inconvenience added?

And as for his explanation as to why this makes sense, I am also rather unconvinced. By that logic, they'd also have to sell the datasheets for their products because the department that writes the documentation is just a cost center, so some incompetent manager could get the idea to close down the department ...

[JPortici]

I use almost exclusively 8bit and 16bit PICs.
XC8 (for PIC) is a propietary compiler, they bought HiTech, only kept the 8bit PIC compiler and its team and keep developing it (the original compiler team is still there.)
XC16 may be GCC but the actual back end may be propietary, as the architecture is propietary. I think the only other compiler is from MikroE and frankly i'd rather stay with unoptimized code, besides the fact that support for new parts and compiler bugs take a huge amount of time to get sorted, as Mr Sanghi correctly pointed out. Again a dedicated team (and they do excellent work).

In these two cases i'm going to "allow" it for the fact that we can use up to -O2 (it's not FREE/STANDARD/PRO anymore, when FREE mean -O0 and PRO was several thousand dollards), in these platforms i'm not going to use -O3 and -Os anyway as i can't really debug it.)
i get my projects done, i use assembly when i need it heavily optimized (not rely on the compiler) and support is responsive even on possible compiler issues even when i write with my gmail account. At least when things get really bad I'm not left to look help in the community (it doesn't seem to work out really well for ST/TI users).

I can absolutely justify it for most of the addons as they seem to be internally developed tools (and they require a separate license, independent on the compiler license)

I can't really justify it for XC8-AVR, XC32 of both sorts. Neither i can for the MISRA check plugin as it's literally a call to CPPCHECK. I asked at CPPCHECK if they were fine with it, received no answer.

All in all i can see the logic in what he said