AMD acquires Xilinx

[Karel]

AMD and Xilinx Stockholders Overwhelmingly Approve AMD’s Acquisition of Xilinx

https://www.amd.com/en/press-releases/2021-04-07-amd-and-xilinx-stockholders-overwhelmingly-approve-amd-s-acquisition

[ataradov]

Intel was ok for Altera, hopefully same holds for Xilinx/AMD.

[RoGeorge]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

[daqq]

Does this type of merges/acquisitions happen in other industries, too?

Yes. See:

[langwadt]

tools

[PKTKS]

Not surprised.

All facts are converging to that since the 90s.

Mostly after 2000s  all video cards fabs. collapsed to just 2 or 3
AMD NV and still trying Intel

All soft small buz failed imploded by MS and or uncle goo...

So having ARM on NV hand  and Xilinx  on AMD and the
full vertical soft stack now almost on hands of just IBM/RHEL
Canonical (MS shell) and goooo Android..

The CARTEL is obviously done and the results are clearly ahead.

Nothing new. CARTEL although "theoretical" disliked..
seem to be very welcome when the money flows on the (proper) hands

Same players.. results will be tragic.. very worried about such moves
repeatedly letting vast majority  hopeless

Paul

[gnuarm]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

Ever hear of a little company called General Motors?  How about Fiat Chrysler Automobiles/Stellantis?  The UK had over a dozen car companies before WWII, but ended up with just two or three, I couldn't keep track.  Other than Rolls and Bentley I'm not sure they have any now... opps, Rolls is owned by BMW and Bentley is owned by VW.  Oh, well.

[JohnnyMalaria]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

Ever hear of a little company called General Motors?  How about Fiat Chrysler Automobiles/Stellantis?  The UK had over a dozen car companies before WWII, but ended up with just two or three, I couldn't keep track.  Other than Rolls and Bentley I'm not sure they have any now... opps, Rolls is owned by BMW and Bentley is owned by VW.  Oh, well.

One of the reasons for the demise of the British car industry was the nationalization of it - British Leyland. They made appalling vehicles. The workers even used to sleep on the job. It became a huge issue in the late 70s. https://www.aronline.co.uk/archive/sleeping-on-the-job-essay/

[JohnnyMalaria]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

In 1993, I joined a company called Glaxo Group Research. In 1995, it merged with Burroughs-Wellcome to become GlaxoWellcome. In another universe, Smith, Kline and French (est. 1830) merged with Beckman Instruments to become SmithKline Beckman and, in turn, with Beecham to become SmithKline Beecham. In 2000, SmithKline Beecham and GlaxoWellcome "merged" (i.e., it was a take-over) to form what is now GlaxoSmithKline. Similarly, in 1998, the Swedish company Astra AB and British company Zeneca merged. Just a couple of examples.

See https://en.wikipedia.org/wiki/List_of_largest_pharmaceutical_mergers_and_acquisitions

[langwadt]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

Ever hear of a little company called General Motors?  How about Fiat Chrysler Automobiles/Stellantis?  The UK had over a dozen car companies before WWII, but ended up with just two or three, I couldn't keep track.  Other than Rolls and Bentley I'm not sure they have any now... opps, Rolls is owned by BMW and Bentley is owned by VW.  Oh, well.

[gnuarm]

Even that graphic doesn't show all the companies that merged into larger companies only to be lost as a brand.  GM had Pontiac and Oldsmobile which no longer exist among others.  AMC was the merger of a number of auto companies and itself is no longer a brand.  Same with many of the British auto makers. 

Largely this is a result of the economies of scale.  We like to gripe about how inefficient large companies are, but in reality size makes a company more efficient.  This is very similar between auto makers and semiconductor companies.  The main difference is the enormous startup costs of forming a new auto company, many billions of dollars.  A semiconductor company can start with much smaller funding because the means of production can be outsourced.  It's still not cheap, but it is manageable.  The only successful new auto company in 50 years is Tesla which is really an EV platform company that markets their products in the form of cars.  What they are really selling is motors and batteries, they just give you the car for free.

[Gyro]

I wonder who's got their sights on Lattice these days - their CEO is ex-AMD and their focus has become quite narrow considering the number of companies they bought up.

[langwadt]

Even that graphic doesn't show all the companies that merged into larger companies only to be lost as a brand.  GM had Pontiac and Oldsmobile which no longer exist among others.  AMC was the merger of a number of auto companies and itself is no longer a brand.  Same with many of the British auto makers. 

and I know since that graphic PSA and FCA have merged

Largely this is a result of the economies of scale.  We like to gripe about how inefficient large companies are, but in reality size makes a company more efficient.  This is very similar between auto makers and semiconductor companies.  The main difference is the enormous startup costs of forming a new auto company, many billions of dollars.  A semiconductor company can start with much smaller funding because the means of production can be outsourced.  It's still not cheap, but it is manageable.  The only successful new auto company in 50 years is Tesla which is really an EV platform company that markets their products in the form of cars.  What they are really selling is motors and batteries, they just give you the car for free.

yeh, when Volkswagen spend bilions making a new car, it'll be sold as a VW, Audi,Seat,Skoda in slightly different packaging but all the internals and the spareparts are the same

[endure]

Toyota built their British manufacturing plant in  Burnaston in 199-92. When  they were advertising jobs they were open to applications from anyone APART from those who had previously worked  in the UK car industry.

[gnuarm]

I wonder who's got their sights on Lattice these days - their CEO is ex-AMD and their focus has become quite narrow considering the number of companies they bought up.

Not sure what you mean.  Lattice has bought up a lot of companies???  I know they bought Silicon Blue quite some time ago.  Or did you mean AMD has bought lots of companies?  How does that indicate a narrow focus?

My main issue with Lattice is while X and A are shooting large bore guns at the high end of FPGAs, Lattice is shooting bird shot at the low end, specifically targeted to the mobile market requiring very tiny packages with very fine PCB features.  No one is addressing a middle range of easy to use, low impact packaging with other than microscope size parts.  I can very seldom use a 1 kLUT device.  I typically need more than 20 or 30 I/Os.  Now days that puts me in a ~200 pin BGA!  I'd like to see a family of FPGAs that encroach on MCU territory. 

I did see where Altera has Cyclone IV parts in a FBGA-169 package with 1.0 mm pitch, but even at JLCPCB it's a $20 part.  That's the smallest part in the family and it has NO multipliers.

[Gyro]

No, Lattice. They bought up Silicon Image too, for their HDMI IP.  They seem to have been hit / constrained by a couple of large stock acquisitions in the past few years and Trump blocked their purchase in 2017 by Canyon Bridge on national security (foreign investment) grounds.

According to Wikipedia, their ex-AMD CEO since 2018 has moved their focus exclusively to low power FPGAs.

https://en.wikipedia.org/wiki/Lattice_Semiconductor

Just a casual personal interest from times past.

[Someone]

My main issue with Lattice is while X and A are shooting large bore guns at the high end of FPGAs, Lattice is shooting bird shot at the low end, specifically targeted to the mobile market requiring very tiny packages with very fine PCB features.  No one is addressing a middle range of easy to use, low impact packaging with other than microscope size parts.  I can very seldom use a 1 kLUT device.  I typically need more than 20 or 30 I/Os.  Now days that puts me in a ~200 pin BGA!  I'd like to see a family of FPGAs that encroach on MCU territory. 

I did see where Altera has Cyclone IV parts in a FBGA-169 package with 1.0 mm pitch, but even at JLCPCB it's a $20 part.  That's the smallest part in the family and it has NO multipliers.

Hand solderable parts? ICE40, MachXO or IGLOO nano, cheap small, stocked.

You want a hand solderable part that has multipliers? Spartan 3, or the odd parts from Efinix, Spartan 6, or ICE40

[Bassman59]

My main issue with Lattice is while X and A are shooting large bore guns at the high end of FPGAs, Lattice is shooting bird shot at the low end, specifically targeted to the mobile market requiring very tiny packages with very fine PCB features.  No one is addressing a middle range of easy to use, low impact packaging with other than microscope size parts.  I can very seldom use a 1 kLUT device.  I typically need more than 20 or 30 I/Os.  Now days that puts me in a ~200 pin BGA!  I'd like to see a family of FPGAs that encroach on MCU territory. 

Microchip (formerly Microsemi, formerly Actel, speaking of acquisitions!) Igloo nano might be the ticket.

I did see where Altera has Cyclone IV parts in a FBGA-169 package with 1.0 mm pitch, but even at JLCPCB it's a $20 part.  That's the smallest part in the family and it has NO multipliers.

$20 without the configuration storage device.

[gnuarm]

No, Lattice. They bought up Silicon Image too, for their HDMI IP.  They seem to have been hit / constrained by a couple of large stock acquisitions in the past few years and Trump blocked their purchase in 2017 by Canyon Bridge on national security (foreign investment) grounds.

According to Wikipedia, their ex-AMD CEO since 2018 has moved their focus exclusively to low power FPGAs.

https://en.wikipedia.org/wiki/Lattice_Semiconductor

Just a casual personal interest from times past.

I think this story has grown with each retelling.  Never take Wikipedia literally without confirming with the sources if they have any.  Going back to the source which is just a report on a Lattice event the only relevant quote is, Lattice wants to "become the low power programmable leader".  They didn't use the much stronger term "moved their focus exclusively to low power FPGAs".  It's not hard to become the low power leader because no one else is really addressing that market.  It correlates with smaller and lower margin devices which the other players are not so interested in.  However, they don't want to wake up one morning to find Lattice and others have eaten their breakfast, so I'm sure they are making inroads into this market.  Not unlike Tesla and the major auto makers.  No one is standing still.  They just aren't falling over themselves to get out of Tesla's way because Tesla has a long way to go to catch up.  Likewise with Lattice.

[gnuarm]

My main issue with Lattice is while X and A are shooting large bore guns at the high end of FPGAs, Lattice is shooting bird shot at the low end, specifically targeted to the mobile market requiring very tiny packages with very fine PCB features.  No one is addressing a middle range of easy to use, low impact packaging with other than microscope size parts.  I can very seldom use a 1 kLUT device.  I typically need more than 20 or 30 I/Os.  Now days that puts me in a ~200 pin BGA!  I'd like to see a family of FPGAs that encroach on MCU territory. 

I did see where Altera has Cyclone IV parts in a FBGA-169 package with 1.0 mm pitch, but even at JLCPCB it's a $20 part.  That's the smallest part in the family and it has NO multipliers.

Hand solderable parts? ICE40, MachXO or IGLOO nano, cheap small, stocked.

You want a hand solderable part that has multipliers? Spartan 3, or the odd parts from Efinix, Spartan 6, or ICE40

I gave a fair description of my needs that very few if any of the Lattice parts meet.  The XO3D does the job, but is not so cheap.  Spartan 3 devices are older than the XP devices that are EOL in the design I may need to refresh.  No way I'm putting them in a new product.  They are also increasing in price. 

I've never found an IGLOO part that was cheap.  Which Spartan 6 part are you talking about?  Maybe the QFP144 that is wider than my board?  The iCE40 parts all either have too few I/Os or too few LUTs.  I have a Lattice selection guide with the useful combinations highlighted and none other than the higher priced XO3D has what I need. 

The Gowin GW1N-9 in a 100 QFP would be perfect.  But they are not well respected by the US government, so it's not a good idea to use their parts in products sold to government agencies.  I don't want to end up in a Huawei like situation. 

[Someone]

My main issue with Lattice is while X and A are shooting large bore guns at the high end of FPGAs, Lattice is shooting bird shot at the low end, specifically targeted to the mobile market requiring very tiny packages with very fine PCB features.  No one is addressing a middle range of easy to use, low impact packaging with other than microscope size parts.  I can very seldom use a 1 kLUT device.  I typically need more than 20 or 30 I/Os.  Now days that puts me in a ~200 pin BGA!  I'd like to see a family of FPGAs that encroach on MCU territory. 

I did see where Altera has Cyclone IV parts in a FBGA-169 package with 1.0 mm pitch, but even at JLCPCB it's a $20 part.  That's the smallest part in the family and it has NO multipliers.

Hand solderable parts? ICE40, MachXO or IGLOO nano, cheap small, stocked.

You want a hand solderable part that has multipliers? Spartan 3, or the odd parts from Efinix, Spartan 6, or ICE40

I gave a fair description of my needs that very few if any of the Lattice parts meet.  The XO3D does the job, but is not so cheap.  Spartan 3 devices are older than the XP devices that are EOL in the design I may need to refresh.  No way I'm putting them in a new product.  They are also increasing in price. 

I've never found an IGLOO part that was cheap.  Which Spartan 6 part are you talking about?  Maybe the QFP144 that is wider than my board?  The iCE40 parts all either have too few I/Os or too few LUTs.  I have a Lattice selection guide with the useful combinations highlighted and none other than the higher priced XO3D has what I need. 

The Gowin GW1N-9 in a 100 QFP would be perfect.  But they are not well respected by the US government, so it's not a good idea to use their parts in products sold to government agencies.  I don't want to end up in a Huawei like situation.

Space, power, price, and performance constrained and you want hand solderable?

I'd like all sorts of things for low volume use, but don't complain that big companies aren't addressing tiny tiny niches.

Small packages need small connections, you're dead in the water right from the start.

p.s. you didn't mention size constraints in the original post.

[gnuarm]

My main issue with Lattice is while X and A are shooting large bore guns at the high end of FPGAs, Lattice is shooting bird shot at the low end, specifically targeted to the mobile market requiring very tiny packages with very fine PCB features.  No one is addressing a middle range of easy to use, low impact packaging with other than microscope size parts.  I can very seldom use a 1 kLUT device.  I typically need more than 20 or 30 I/Os.  Now days that puts me in a ~200 pin BGA!  I'd like to see a family of FPGAs that encroach on MCU territory. 

I did see where Altera has Cyclone IV parts in a FBGA-169 package with 1.0 mm pitch, but even at JLCPCB it's a $20 part.  That's the smallest part in the family and it has NO multipliers.

Hand solderable parts? ICE40, MachXO or IGLOO nano, cheap small, stocked.

You want a hand solderable part that has multipliers? Spartan 3, or the odd parts from Efinix, Spartan 6, or ICE40

I gave a fair description of my needs that very few if any of the Lattice parts meet.  The XO3D does the job, but is not so cheap.  Spartan 3 devices are older than the XP devices that are EOL in the design I may need to refresh.  No way I'm putting them in a new product.  They are also increasing in price. 

I've never found an IGLOO part that was cheap.  Which Spartan 6 part are you talking about?  Maybe the QFP144 that is wider than my board?  The iCE40 parts all either have too few I/Os or too few LUTs.  I have a Lattice selection guide with the useful combinations highlighted and none other than the higher priced XO3D has what I need. 

The Gowin GW1N-9 in a 100 QFP would be perfect.  But they are not well respected by the US government, so it's not a good idea to use their parts in products sold to government agencies.  I don't want to end up in a Huawei like situation.

Space, power, price, and performance constrained and you want hand solderable?

I never said anything about hand solderable.  That was you. 

I'd like all sorts of things for low volume use, but don't complain that big companies aren't addressing tiny tiny niches.

Small packages need small connections, you're dead in the water right from the start.

p.s. you didn't mention size constraints in the original post.

I also didn't give my address and phone number.  I described what I need in general terms as being the mid range without the large BGA packaging and without the micro requirements on board design rules.  People have told me more than once it's not hard to find vendors that can work with 0.5 mm pitch BGAs, but they are still high pin count and so more expensive.  They test every I/O and that takes time.  Test time on the very expensive testers are what separates the low cost chips from the more expensive parts until the die gets big enough to dominate the costs. 

[jmelson]

$20 without the configuration storage device.

Spartan 3A in the 50K LUT size is $10 without config device, or $13 with internal SPROM.  I use the SST (now Altera, I think) serial EPROMS for external loading, they are something like $0.79.

Jon

[srb1954]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

Ever hear of a little company called General Motors?  How about Fiat Chrysler Automobiles/Stellantis?  The UK had over a dozen car companies before WWII, but ended up with just two or three, I couldn't keep track.  Other than Rolls and Bentley I'm not sure they have any now... opps, Rolls is owned by BMW and Bentley is owned by VW.  Oh, well.

Morgan is still independent and Aston-Martin is semi-independent of the global conglomerates.

[EEEnthusiast]

Too many mergers is bad for engineering jobs.. Every time they merge, they identify redundant jobs and fire the engineers who are doing the same jobs in both companies.

[peter-h]

I used to do FPGA designs and did a bit of that and ASIC design consultancy in the 1990s, and wondered just how far FPGAs can go before they become really marginal.

They were great when CPUs were slow (4MHz) and I used the X2000 and X3000 families effectively. But the tools (Viewlogic 4 and XACT5) were horrible, not generally backwards compatible, the company was arrogant about that (the salesman said "got to crack some eggs to make omelletes") so I had to archive the old tools (nowadays you could probably use VMWARE), had to get cracks for their dongles (if they broke they left you hanging, and I had two of them concurrently, and I doubt they would run under VMWARE).

Today you get a $7 ARM which runs at 168MHz and every one of the designs I did back then could be done as a state machine from an interrupt running at say 10-100kHz. And that's without doing anything clever.

I was really glad to get out of the FPGA business ~ 20 years ago. Since then I have seen their capacities go sky high, so big that nobody can design anything in them unless they either drop in large blocks (called "IP" these days), drop in RAM arrays, drop in a 128 bit barrel shifter or is a wizard in a hardware description language (VHDL was used way back). And their prices went sky-high. Well, they were always high - except for the bottom-end devices. The bigger X4000 devices were £100+ and thus with very limited applications.

The remaining obvious applications were things like a 20 channel ARINC429 + 1553 interface card (with FIFOs etc) which would cost a fortune if built with chips from the likes of HOLTEK, and with an FPGA it costs a smaller fortune And crypto (RSA and such) and packet routing for high speed links. These applications will always exist but the volumes aren't going to be great.

[NiHaoMike]

The cost of making ASICs using a modern process is high enough that the volume would have to be really large and/or the gate count very high to justify going that route instead of FPGAs. As an example, some very early Japanese market Prius used FPGAs (likely sharing many design elements with the RAV4 EV at the time), then they switched to ASICs since there was a huge cost savings. But lately, they're going back to FPGAs for their hybrid and fuel cell vehicles instead of designing new ASICs since modern FPGAs are a lot cheaper so going to ASICs would not save much.

[peter-h]

Sure, but how much does a car cost?

The 4 digit product market is not enough to support a company of Xilinx's size into the future.

FPGAs have been widely used for ASIC prototyping and then the device cost is totally irrelevant, but it is a very small % of the FPGA market.

[langwadt]

The cost of making ASICs using a modern process is high enough that the volume would have to be really large and/or the gate count very high to justify going that route instead of FPGAs. As an example, some very early Japanese market Prius used FPGAs (likely sharing many design elements with the RAV4 EV at the time), then they switched to ASICs since there was a huge cost savings. But lately, they're going back to FPGAs for their hybrid and fuel cell vehicles instead of designing new ASICs since modern FPGAs are a lot cheaper so going to ASICs would not save much.

afaik cellphone basestations was (is?) a huge market for FPGAs

[gnuarm]

The cost of making ASICs using a modern process is high enough that the volume would have to be really large and/or the gate count very high to justify going that route instead of FPGAs. As an example, some very early Japanese market Prius used FPGAs (likely sharing many design elements with the RAV4 EV at the time), then they switched to ASICs since there was a huge cost savings. But lately, they're going back to FPGAs for their hybrid and fuel cell vehicles instead of designing new ASICs since modern FPGAs are a lot cheaper so going to ASICs would not save much.

I'm curious as to what in an EV requires an FPGA rather than an MCU or DSP?  The calculations are not so intense or high speed.  I am a big proponent of using FPGAs for such control applications, but most of the community does not agree with me.  Where did you read about this?

afaik cellphone basestations was (is?) a huge market for FPGAs

Yes, in fact, comms in general are the big driver for the large FPGAs especially.  You gotta wonder what anyone is doing with 8 million LUTs, but they make it because they can sell it.

[Bassman59]

The cost of making ASICs using a modern process is high enough that the volume would have to be really large and/or the gate count very high to justify going that route instead of FPGAs. As an example, some very early Japanese market Prius used FPGAs (likely sharing many design elements with the RAV4 EV at the time), then they switched to ASICs since there was a huge cost savings. But lately, they're going back to FPGAs for their hybrid and fuel cell vehicles instead of designing new ASICs since modern FPGAs are a lot cheaper so going to ASICs would not save much.

I'm curious as to what in an EV requires an FPGA rather than an MCU or DSP?  The calculations are not so intense or high speed.  I am a big proponent of using FPGAs for such control applications, but most of the community does not agree with me.  Where did you read about this?

I have no way of knowing whether NiHaoMike's assertions are correct, but the one thing the FPGA gives over an MCU or DSP is flexibility. How many SPI peripherals do you need? The MCU has a limit. Maybe 8 on that NXP LPC55 thing with its FLEXCOMMs. In an FPGA? How many pins ya got?

It may very well be that those FPGAs in the car have an embedded processor doing processor-type stuff and they add on the exact peripherals needed.

[gnuarm]

I'm curious as to what in an EV requires an FPGA rather than an MCU or DSP?  The calculations are not so intense or high speed.  I am a big proponent of using FPGAs for such control applications, but most of the community does not agree with me.  Where did you read about this?

I have no way of knowing whether NiHaoMike's assertions are correct, but the one thing the FPGA gives over an MCU or DSP is flexibility. How many SPI peripherals do you need? The MCU has a limit. Maybe 8 on that NXP LPC55 thing with its FLEXCOMMs. In an FPGA? How many pins ya got?

It may very well be that those FPGAs in the car have an embedded processor doing processor-type stuff and they add on the exact peripherals needed.

That's my point.  There is no use in idle speculation.  We might as well imagine they are designing with 7400 TTL or hand wired transistors.  I saw a computer built from transistors in high school.  Our electronics class was given one from the weather service.  A FF was a small board.  Our instructor showed us the soldering and pointed out that flaws we could not see would flunk it for military or space applications.  I believe it used a magnetic drum for main memory.  Much of the circuitry was likely just for timing the memory accesses. 

A car could do something similar since it has so many rotating parts.  lol  See where speculation gets us?

[Wolfram]

The cost of making ASICs using a modern process is high enough that the volume would have to be really large and/or the gate count very high to justify going that route instead of FPGAs. As an example, some very early Japanese market Prius used FPGAs (likely sharing many design elements with the RAV4 EV at the time), then they switched to ASICs since there was a huge cost savings. But lately, they're going back to FPGAs for their hybrid and fuel cell vehicles instead of designing new ASICs since modern FPGAs are a lot cheaper so going to ASICs would not save much.

I'm curious as to what in an EV requires an FPGA rather than an MCU or DSP?  The calculations are not so intense or high speed.  I am a big proponent of using FPGAs for such control applications, but most of the community does not agree with me.  Where did you read about this?

I can only speculate on the justification, but Tesla drive units I've taken apart have used FPGAs as part of the motor control. The big rear drive unit in the Model S uses an Actel/Microsemi ProASIC3 for example, along with a couple of TI motor control DSPs.

[peter-h]

Most motor control today is 3 phase brushless stuff and most of it is done with fairly fast CPUs e.g. ARM32F. These have 1-2us conversion time 12-bit ADCs, 12 bit DACs, and are easily capable of doing this job to any level of sophistication.

The problem is that designers like job security and there is nothing better than an FPGA for that. Nobody can do anything with it once the original designer is gone

[filssavi]

Most motor control today is 3 phase brushless stuff and most of it is done with fairly fast CPUs e.g. ARM32F. These have 1-2us conversion time 12-bit ADCs, 12 bit DACs, and are easily capable of doing this job to any level of sophistication.

The problem is that designers like job security and there is nothing better than an FPGA for that. Nobody can do anything with it once the original designer is gone

That is as long as you are unsina a VSI inverter with standard modulation, or maybe a diode clamped NPC (if you chose the correct MCU) anything more than that and the PWM modulators/timers will not be sufficient anymore
Than you are forced to go to an FPGA  just to have enough of the right kind of PWM channels ( most STM32 while more than capable enough from a computational power perspective have only 1 “advanced” timer that does deadtime, complementary outputs etc.

[langwadt]

Most motor control today is 3 phase brushless stuff and most of it is done with fairly fast CPUs e.g. ARM32F. These have 1-2us conversion time 12-bit ADCs, 12 bit DACs, and are easily capable of doing this job to any level of sophistication.

The problem is that designers like job security and there is nothing better than an FPGA for that. Nobody can do anything with it once the original designer is gone

when you have 400V and 1000A it makes sense to have some hardware as a last defense against the software doing something silly

[langwadt]

Most motor control today is 3 phase brushless stuff and most of it is done with fairly fast CPUs e.g. ARM32F. These have 1-2us conversion time 12-bit ADCs, 12 bit DACs, and are easily capable of doing this job to any level of sophistication.

The problem is that designers like job security and there is nothing better than an FPGA for that. Nobody can do anything with it once the original designer is gone

That is as long as you are unsina a VSI inverter with standard modulation, or maybe a diode clamped NPC (if you chose the correct MCU) anything more than that and the PWM modulators/timers will not be sufficient anymore
Than you are forced to go to an FPGA  just to have enough of the right kind of PWM channels ( most STM32 while more than capable enough from a computational power perspective have only 1 “advanced” timer that does deadtime, complementary outputs etc.

doesn't most if not all F4/F7 have two advanced timers?

[filssavi]

doesn't most if not all F4/F7 have two advanced timers?

I am pretty sure ( not 1000% though) that only selected parts in the F0 F3/G4 and H7 lines have the motor control timers  (no F4) the top of the line G4 has 3 advanced timers, that is 18 pwm and 12 additional from the HR timer  that is 30 PWMs
That is however no many in the big schema of thing, a colleague of mine was working on a system where the controller was responsible for. 4 NPC converters and that is 48 PWM right there.

Also the total determinism of FPGAs and CPLDs is very handy in making sure that nothing blows up for unexpected reasons.
And it makes safety certifications such as SIL much easier.

All in all it comes down to the fact that even a 100€ FPGA is not that expensive of a BOM item in a converter the size of a small room where you are paying multiple hundreds of dollars (or even thousands) in capacitor alone

[langwadt]

doesn't most if not all F4/F7 have two advanced timers?

I am pretty sure ( not 1000% though) that only selected parts in the F0 F3/G4 and H7 lines have the motor control timers  (no F4)

I've mostly used the F407 it has two advanced timers with complemenetary outputs and programmable dead time

[peter-h]

I can tell you that "total determinism of FPGAs" is fiction

It is close, if designed correctly, paying lots of attention to sync clocking etc, but it is easy to generate a design which is temperature-marginal, metastable, etc.

There is no fundamental difference between complex logic in hardware, and software. There are just different ways to screw up

FPGAs are ever so tempting for many projects because you can go for a "direct solution" whereas with software you often have to use imagination (clever use of timers, etc). But you pay for that dead-end development route afterwards, with dead-end tools which go unsupported after a year or two, and a project nobody can pick up.

I don't know what the current Xilinx tool licensing is (it isn't dongles anymore) but if they use FlexLM that is a nightmare.

[gnuarm]

I can tell you that "total determinism of FPGAs" is fiction

It is close, if designed correctly, paying lots of attention to sync clocking etc, but it is easy to generate a design which is temperature-marginal, metastable, etc.

There is no fundamental difference between complex logic in hardware, and software. There are just different ways to screw up

Maybe you have been smoking a bit too much of the wacky weed?  I literally have no idea what you are talking about.  Metastability is so easy to avoid.  In fact, it is often not even a problem to begin with unless your design is running close to the limit of the FPGA delays on the path of interest.

i don't know what you are talking about temperature marginal.  You mean the timing fails when the device runs warm?  There is no reason for that unless the tools fail to properly analyze timing... or you just don't run timing analysis. 

Anyone can be a bad designer I suppose is what you are saying.  My experience is that with MCUs designers attempt to manage potential failures, but often they are not recognized up front.  With HDL the various failures (other than just not designing the right logic) are much fewer and easier to avoid.  Even errors in logic are not hard to find if you do a proper job of validating the design.

FPGAs are ever so tempting for many projects because you can go for a "direct solution" whereas with software you often have to use imagination (clever use of timers, etc). But you pay for that dead-end development route afterwards, with dead-end tools which go unsupported after a year or two, and a project nobody can pick up.

I don't know what the current Xilinx tool licensing is (it isn't dongles anymore) but if they use FlexLM that is a nightmare.

Again, I have no idea what you mean about "dead-end tools".  HDL is not tied to a specific tool.  Just like compilers synthesis software isolates the language from the target.  There is little reason today to write code with special features that require specific tools for implementation. 

I just don't know what your issues with FPGAs and the tools are based on, but it must have been some horrific adventures at some point in the past.

[daqq]

FPGAs are ever so tempting for many projects because you can go for a "direct solution" whereas with software you often have to use imagination (clever use of timers, etc). But you pay for that dead-end development route afterwards, with dead-end tools which go unsupported after a year or two, and a project nobody can pick up.

Are you quite sure about that? HDL is far more portable than 'cleverly used timers' between various vendors. And please note that 'using your imagination' occasionally translates into 'Create a Rube Goldbergesque mechanism where the five timers are used in conjunction with the DMA, USART and SPI set to an obscure setting and connected externally'.

As to unsupported tools... where? You can download up to date tools that will open ancient projects or download the ancient tools. Sure, not every tool in FPGAland is perfect and Xilinx are a bit dickish about updating their tools for their older devices (meaning series 6 since series 7 came out ~10 years ago), but there are other vendors out there who have the equivalent stuff.

edit: And dropping support for a tool after a year or two after it came out has been my experience with MCUs rather than FPGAs. I'm sure that it can happen, but it's rare on both fronts.

Could you please tell me what kind of project traumatised you to this level against FPGAs? Was it recent or are you just comparing the state of the FPGA tools from 20 years ago to the state of the MCU tools today?

[filssavi]

I can tell you that "total determinism of FPGAs" is fiction

It is close, if designed correctly, paying lots of attention to sync clocking etc, but it is easy to generate a design which is temperature-marginal, metastable, etc.

There is no fundamental difference between complex logic in hardware, and software. There are just different ways to screw up

FPGAs are ever so tempting for many projects because you can go for a "direct solution" whereas with software you often have to use imagination (clever use of timers, etc). But you pay for that dead-end development route afterwards, with dead-end tools which go unsupported after a year or two, and a project nobody can pick up.

I don't know what the current Xilinx tool licensing is (it isn't dongles anymore) but if they use FlexLM that is a nightmare.

Even the free version of vivado (I dont know intel tools but i suspect it it the same)  allows you (and it is actually configured by default) to perform timing analisys in all corners of the PVT space, so if you use the tool correctly it will find the timing violations. Of course if the design is poor with wierd unconstrained CDCs or large use of asynchronous logic for example you might have timing problems, but that is neither the FPGA nor the tools fault.

Now if you put softcores inside the FPGA, you might have some non deterministic behaviour if the code to logic interface is not extremely well designed, but that is not the logic's fault

[TMM]

Wow!  Looks like soon all the chip manufacturers will become a single company.

Does this type of merges/acquisitions happen in other industries, too?

Ever hear of a little company called General Motors?  How about Fiat Chrysler Automobiles/Stellantis?  The UK had over a dozen car companies before WWII, but ended up with just two or three, I couldn't keep track.  Other than Rolls and Bentley I'm not sure they have any now... opps, Rolls is owned by BMW and Bentley is owned by VW.  Oh, well.

Morgan is still independent and Aston-Martin is semi-independent of the global conglomerates.

Mercedes-Benz is set to own 20% of Aston Martin, and the cars already use Mercedes-Benz engines/driveline and infotainment among other things if I'm not mistaken. Pretty inevitable in an age where cars are becoming a smartphone on wheels and the profit margins are ever smaller, even on a luxury car.

[gnuarm]

Mercedes-Benz is set to own 20% of Aston Martin, and the cars already use Mercedes-Benz engines/driveline and infotainment among other things if I'm not mistaken. Pretty inevitable in an age where cars are becoming a smartphone on wheels and the profit margins are ever smaller, even on a luxury car.

You don't know the half of it.  I have a Tesla model X and the durn thing is worse than my cell phone with features constantly appearing and disappearing, UI changes and just plain not working very well since nearly everything in the car is a "beta" feature.  I think my cell phone is actually less disruptive.  At least I can google to find how to fix the issues in my phone.  I can google from the car (even if it is slow and clumsy) but there are never fixes, just commiseration.

[JohnnyMalaria]

Mercedes-Benz is set to own 20% of Aston Martin, and the cars already use Mercedes-Benz engines/driveline and infotainment among other things if I'm not mistaken. Pretty inevitable in an age where cars are becoming a smartphone on wheels and the profit margins are ever smaller, even on a luxury car.

You don't know the half of it.  I have a Tesla model X and the durn thing is worse than my cell phone with features constantly appearing and disappearing, UI changes and just plain not working very well since nearly everything in the car is a "beta" feature.  I think my cell phone is actually less disruptive.  At least I can google to find how to fix the issues in my phone.  I can google from the car (even if it is slow and clumsy) but there are never fixes, just commiseration.

A few years ago, my wife bought a new Mini Cooper but declined Mini's equivalent of On-Star. A few months later, we went out to eat. When we got back to the car, it was raining and the windshield wipers wouldn't work. They had been about an hour earlier. I half-jokingly said that Mini must have pushed a firmware update. She called the dealership and that is exactly what happened. So even though she didn't have the On-Star-type service, the car was still in regular communication with Mini.

BTW, if you ever get the chance, take the 3+ hour tour of the Mini manufacturing plant in the UK - an amazing experience.

[asmi]

I know that FPGAs and FPGA+CPU SoCs are used a lot in ADAS systems. If you have adaptive cruise in your car, or 360° camera, odds are there is a Zynq hiding somewhere doing the heavy lifting

[Bud]

I have a Tesla model X and the durn thing is worse than my cell phone with features constantly appearing and disappearing, UI changes and just plain not working very well since nearly everything in the car is a "beta" feature.

Wow, that is plain dangerous. They should not be doing that type of experimentation in the field. Would certainly drive me mad.

[gnuarm]

I have a Tesla model X and the durn thing is worse than my cell phone with features constantly appearing and disappearing, UI changes and just plain not working very well since nearly everything in the car is a "beta" feature.

Wow, that is plain dangerous. They should not be doing that type of experimentation in the field. Would certainly drive me mad.

It is nice that they continually improve the car in many ways... all free.  But it's not so fun knowing that at any time features can change or even not work as well.  Mean while they have never gotten the automatic high beam control and the automatic wipers to work fully.  It's not like they designed these features last year.  After 8 years these things still work very marginally.  For the head lights I can take full manual control.  For the wipers the choices are automatic with manual override, or no intermittent wiper at all.  They don't have a knob on the wiper stalk to let you dial in the delay rate.  I guess they figured they would be able to make it work well.

[srb1954]

Morgan is still independent and Aston-Martin is semi-independent of the global conglomerates.

Mercedes-Benz is set to own 20% of Aston Martin, and the cars already use Mercedes-Benz engines/driveline and infotainment among other things if I'm not mistaken. Pretty inevitable in an age where cars are becoming a smartphone on wheels and the profit margins are ever smaller, even on a luxury car.

Yes, Mercedes is extending its tentacles into Aston-Martin but they don't have a controlling interest over the company which is why I would describe Aston-Martin as still being semi-independent. The increasing usage of brought-in drive trains and infotainment systems is inevitable these days for small volume manufacturers as they don't have the financial and engineering resources to develop the very complex equipment demanded by today's market.

By the way Morgan have long used BMW drive trains in their cars but they still continue on their independent path with their idiosyncratic car designs.

[gnuarm]

In other news, AMD took a big step up today because a new analyst gave it a rating of "outperform".  Up nearly 6% dragging XLNX with it.  I own some Xilinx stock and find it funny that it trades consistently about $10-$14 below what it will be worth when the stock trade is consummated.  The share holders have blessed the deal, so I guess the only worry is the uncertainty of the AMD stock price at the point of closing.  Seems odd to me the gap is not narrowing any. 

[langwadt]

Morgan is still independent and Aston-Martin is semi-independent of the global conglomerates.

Mercedes-Benz is set to own 20% of Aston Martin, and the cars already use Mercedes-Benz engines/driveline and infotainment among other things if I'm not mistaken. Pretty inevitable in an age where cars are becoming a smartphone on wheels and the profit margins are ever smaller, even on a luxury car.

Yes, Mercedes is extending its tentacles into Aston-Martin but they don't have a controlling interest over the company which is why I would describe Aston-Martin as still being semi-independent. The increasing usage of brought-in drive trains and infotainment systems is inevitable these days for small volume manufacturers as they don't have the financial and engineering resources to develop the very complex equipment demanded by today's market.

By the way Morgan have long used BMW drive trains in their cars but they still continue on their independent path with their idiosyncratic car designs.

Aston Martin has used Mercedes engines and other parts for a number of years now, before that they were owned by Ford. And you can clearly see that some Fords and Astons were done or at least inspired by by the same designer

[tszaboo]

I have a Tesla model X and the durn thing is worse than my cell phone with features constantly appearing and disappearing, UI changes and just plain not working very well since nearly everything in the car is a "beta" feature.

Wow, that is plain dangerous. They should not be doing that type of experimentation in the field. Would certainly drive me mad.

It is nice that they continually improve the car in many ways... all free.  But it's not so fun knowing that at any time features can change or even not work as well.  Mean while they have never gotten the automatic high beam control and the automatic wipers to work fully.  It's not like they designed these features last year.  After 8 years these things still work very marginally.  For the head lights I can take full manual control.  For the wipers the choices are automatic with manual override, or no intermittent wiper at all.  They don't have a knob on the wiper stalk to let you dial in the delay rate.  I guess they figured they would be able to make it work well.

Beams still not working properly? About 4 years ago, I was seriously missing a pump action shotgun from my car (unfortunately illegal in Europe) because a Tesla was driving behind me for about 20 minutes on a mountain road at night.
The company feels like just an experiment. With surprise bills for repairs, cars falling apart because they forget bolts, and most important features are fart sound easter eggs.

[Bassman59]

In other news, AMD took a big step up today because a new analyst gave it a rating of "outperform".  Up nearly 6% dragging XLNX with it.  I own some Xilinx stock and find it funny that it trades consistently about $10-$14 below what it will be worth when the stock trade is consummated.  The share holders have blessed the deal, so I guess the only worry is the uncertainty of the AMD stock price at the point of closing.  Seems odd to me the gap is not narrowing any.

I own both XLNX and AMD, and the other day I checked the XLNX price to see if it was near the tender offer price, and I was surprised to see it was not. (If it was, I was going to sell it.) So I'll just keep an eye on it. If it does hit or exceed the offer price, I'll sell. I remember that for the longest time I had TEK shares, and when they were bought by Fluke I figured I'd just let Fluke send me the money. Well, it turns out there was some bullshit fee associated with that, and the fee was greater than the commission had I sold. (Now of course there are no more commissions.)

Several years ago when NXP was on the block, I had shares, and I kept watching the news and the ticker price. The shares did reach the offer price, but there was concern that the sale would not go through, so I sold the shares and then when the deal collapsed the NXP price went down too.

[Bassman59]

i don't know what you are talking about temperature marginal.  You mean the timing fails when the device runs warm?  There is no reason for that unless the tools fail to properly analyze timing... or you just don't run timing analysis.

Funny, yesterday morning the boss came in and said, "something weird happened yesterday. I was doing some testing before buttoning-up the thing, and it was giving weird data. At first I thought, did you change the firmware?" Nope, same firmware that's been in it for a few weeks now, I'm gonna release it soon.

Turns out that the key words above are "before buttoning-up." In a proper installation the board with big FPGA is installed such that you put a silpad between the chip and the big cold finger the chip goes up against when the board is installed. if the silpad isn't there and the board isn't screwed into place, the FPGA gets warm. Warm enough to no longer meet timing.

[gnuarm]

Turns out that the key words above are "before buttoning-up." In a proper installation the board with big FPGA is installed such that you put a silpad between the chip and the big cold finger the chip goes up against when the board is installed. if the silpad isn't there and the board isn't screwed into place, the FPGA gets warm. Warm enough to no longer meet timing.

Sure, the temperature specs on FPGAs are junction temperature.  If you are running commercial temp range, if the junction gets out of that range the timing is not guaranteed.  No different from running your car with no coolant.  Not the car's fault and not the FPGA or tools fault.

20 years ago I was working on an Altera 10K design (I think that's what the chips were called).  It was actually an upgrade to test gear to add ATM, if I recall.  When they first worked that design they had trouble with timing as there was some disconnect in the tools or our timing constraints.  Altera would not help us with it so in the respin of the FPGA we had to do our own temperature qualification of the design.  We ended up running many iterations of the MAX+II tools with different seeds every night and testing them until we found one that worked.  They were planning to discontinue the tool, I think they didn't want to find any significant bugs that would need to be fixed.  They may have brought back the MAX+II tool later on, not sure why.

That was the only time in 25 years of working with FPGAs that I've seen that sort of a problem with the tools. 

[Bassman59]

Turns out that the key words above are "before buttoning-up." In a proper installation the board with big FPGA is installed such that you put a silpad between the chip and the big cold finger the chip goes up against when the board is installed. if the silpad isn't there and the board isn't screwed into place, the FPGA gets warm. Warm enough to no longer meet timing.

Sure, the temperature specs on FPGAs are junction temperature.  If you are running commercial temp range, if the junction gets out of that range the timing is not guaranteed.  No different from running your car with no coolant.  Not the car's fault and not the FPGA or tools fault.

Right .. remember the tools tell you whether the design will work over the stated temperature/voltage range.

They don't tell you whether your design will or will not work outside of the range. See, it might, or it might not ... so, you must assume that it won't.

And we proved that, hey, yeah, when it's outside of the temperature spec, it doesn't work.

And further is that you can never tell exactly what won't work outside of spec range. Do you feel lucky?

This is why I never overclock a CPU.