Speculation: MME Driver Monitoring System powered by Veoneer HW and Seeing Machines logic

macchiaz-o

Well-Known Member
First Name
Jonathan
Joined
Nov 25, 2019
Messages
3,234
Reaction score
6,537
Location
Valley of the Sun
First Name
Jonathan
Vehicles
human driven
Country flag
From Colin Barnden, Principal Analyst, Semicast Research:

More #detectivework. The "global US-HQd auto OEM" can only conceivably be Ford Motor Company, based on press announcements. We know Fovio from Seeing Machines is a Xilinx Zynq-7000 FPGA and have established that the profile of the DMS in the Mustang Mach-E is identical to that offered by Veoneer. Hence am calling Ford's "Active Drive Assist" as Veoneer/Xilinx/Seeing Machines. What interests me is the update coming over-the-air (OTA) in 2021. https:/lInkd. in/dUNY-gJ We know Ford is using Mobileye EyeQ4 in Active Drive Assist. https://Inkd. in/dNtTC88 EyeQ4 is a fixed-function chip so the only conclusion can come to is that the 0TA update in 2021 is for the #DMS and probably the production-ready version of the newly announced Occula core. https://Inkd.in/evraTzU So in 2021 am speculating Ford might have by far the most sophisticated DMS on the market in the Mach-E and maybe F-150. That would be an incredible achievement for the Blue Oval. We'll know more next year...

image_58f3691a-8150-405a-9fb5-10f3556f5d5920200921_075857.jpg





Advertisement

 

RyZt

Well-Known Member
Joined
Nov 17, 2019
Messages
441
Reaction score
495
Location
San Jose
Vehicles
Mach E4X, previously CMax Hybrid
Country flag
> "two premium German OEM" "four OEMs with numerous new vehicle models launching in 2020-2022 time frame"

So we're looking at Active Drive Assist from 4 automakers: GM, Ford, VW/Audi/Porsche, BMW.

> Fovio from Seeing Machines is a Xilinx Zynq-7000 FPGA

Back when I was in school, I learned that FPGA is for development whereas ASIC is used for production products. The former allows re-programming, but is a lot more expensive than the latter.

In the world of OTA, is FPGA being shipped to end customers for production use nowadays?
 
Last edited:

RonTCat

Well-Known Member
First Name
Ron
Joined
Jul 21, 2020
Messages
710
Reaction score
1,281
Location
USA
First Name
Ron
Vehicles
Mach-E wannabuy
Country flag
> "two premium German OEM" "four OEMs with numerous new vehicle models launching in 2020-2022 time frame"

So we're looking at Active Drive Assist from 4 automakers: GM, Ford, VW/Audi/Porsche, BMW.

> Fovio from Seeing Machines is a Xilinx Zynq-7000 FPGA

Back when I was in school, I learned that FPGA is for development whereas ASIC is used for production products. The former allows re-programming, but is a lot more expensive than the latter.

In the world of OTA, is FPGA being shipped to end customers for production use nowadays?
I think FPGA prices have dropped to the point of blurring the line. You should also be able to get reprogrammable ASIC.
 
OP
macchiaz-o

macchiaz-o

Well-Known Member
First Name
Jonathan
Joined
Nov 25, 2019
Messages
3,234
Reaction score
6,537
Location
Valley of the Sun
First Name
Jonathan
Vehicles
human driven
Country flag
  • Thread starter
  • Thread Starter
  • #4
I think FPGA prices have dropped to the point of blurring the line. You should also be able to get reprogrammable ASIC.
The Xilinx chip used here is a SoC containing a dual core ARM processor and an FPGA. Updates to code for the general purpose processor (the CPU) are a lot easier to prepare and less space constrained than updates to the logic/IP loaded into the FPGA. But the FPGA can run its loaded design much more efficiently (i.e. handling much higher throughput of raw I/O) than the CPU.

Both are more readily upgradable over a typical ASIC. It takes a few seconds at most to load a new logic design to the FPGA and maybe a little longer to load the ARM side. It may not be cheap, but it definitely meets the requirements for updateability for adding future fixes and enhancements. This would be way more constrained if they chose an ASIC.
 

RonTCat

Well-Known Member
First Name
Ron
Joined
Jul 21, 2020
Messages
710
Reaction score
1,281
Location
USA
First Name
Ron
Vehicles
Mach-E wannabuy
Country flag
The Xilinx chip used here is a SoC containing a dual core ARM processor and an FPGA. Updates to code for the general purpose processor (the CPU) are a lot easier to prepare and less space constrained than updates to the logic/IP loaded into the FPGA. But the FPGA can run its loaded design much more efficiently (i.e. handling much higher throughput of raw I/O) than the CPU.

Both are more readily upgradable over a typical ASIC. It takes a few seconds at most to load a new logic design to the FPGA and maybe a little longer to load the ARM side. It may not be cheap, but it definitely meets the requirements for updateability for adding future fixes and enhancements. This would be way more constrained if they chose an ASIC.
What he said. FPGA are preferred by high speed trading companies, where raw I/O performance is king. This application likely benefits from the same thing.
 

GoGoGadgetMachE

Well-Known Member
First Name
Michael
Joined
Jan 23, 2020
Messages
2,882
Reaction score
5,823
Location
Ohio
First Name
Michael
Vehicles
2013 Ford Fusion Energi, 2021 Mach-E First Edition (ordered)
Country flag
as a reference point, it's public knowledge that at least Google and Microsoft both use FPGAs (e.g. Microsoft offloads networking to FPGA-powered NICs • The Register (theregister.com) ) for mainstream-ish use cases and have for years. "Cloud computing" has pushed the industry forward pretty substantially here, as has finance (at @RonTCat correctly points out).

The more interesting thing to me here is that this is the tech rather than some Nvidia offering, especially with Nvidia likely buying ARM, and Nvidia's growing automotive presence, it's an interesting time for the industry.
 
OP
macchiaz-o

macchiaz-o

Well-Known Member
First Name
Jonathan
Joined
Nov 25, 2019
Messages
3,234
Reaction score
6,537
Location
Valley of the Sun
First Name
Jonathan
Vehicles
human driven
Country flag
  • Thread starter
  • Thread Starter
  • #7
as a reference point, it's public knowledge that at least Google and Microsoft both use FPGAs (e.g. Microsoft offloads networking to FPGA-powered NICs • The Register (theregister.com) ) for mainstream-ish use cases and have for years. "Cloud computing" has pushed the industry forward pretty substantially here, as has finance (at @RonTCat correctly points out).

The more interesting thing to me here is that this is the tech rather than some Nvidia offering, especially with Nvidia likely buying ARM, and Nvidia's growing automotive presence, it's an interesting time for the industry.
Yeah the whole Nvidia buying ARM thing is interesting. ARM wasn't a manufacturer of chips -- they only licensed their chip designs, I think. But Nvidia does make chips. So I wonder how this will eventually impact other companies that license their designs, like Qualcomm, Samsung, Apple, and many others.

Back to the Mach-E DMS: I could be wrong here, but I don't think Nvidia makes any FPGAs at all. Their focus is on GPUs, and to a far lesser degree, CPUs.

FPGAs and GPUs are wholly different beasts. If Seeing Machines' expertise in DMS relies upon their logic design IP that they've set up to run in an FPGA, porting it over to run efficiently on a GPU or CPU wouldn't be an easy or inexpensive task (IMO). And it'd most likely come at a high runtime cost too, in terms of cooling, board bandwidth, and power requirements.

I should caveat everything I've said though -- my experience is in "application development" which is just regular computer programming like on servers and desktops. I don't have any personal experience writing GPU code or FPGA logic, but I have worked (only a little bit) with people who do that type of work.
 

GoGoGadgetMachE

Well-Known Member
First Name
Michael
Joined
Jan 23, 2020
Messages
2,882
Reaction score
5,823
Location
Ohio
First Name
Michael
Vehicles
2013 Ford Fusion Energi, 2021 Mach-E First Edition (ordered)
Country flag
Yeah the whole Nvidia buying ARM thing is in thteresting. ARM wasn't a manufacturer of chips -- they only licensed their chip designs, I think. But Nvidia does make chips. So I wonder how this will eventually impact other companies that license their designs, like Qualcomm, Samsung, Apple, and many others.

Back to the Mach-E DMS: I could be wrong here, but I don't think Nvidia makes any FPGAs at all. Their focus is on GPUs, and to a far lesser degree, CPUs.

FPGAs and GPUs are wholly different beasts. If Seeing Machines' expertise in DMS relies upon their logic design IP that they've set up to run in an FPGA, porting it over to run efficiently on a GPU or CPU wouldn't be an easy or inexpensive task (IMO). And it'd most likely come at a high runtime cost too, in terms of cooling, board bandwidth, and power requirements.

I should caveat everything I've said though -- my experience is in "application development" which is just regular computer programming like on servers and desktops. I don't have any personal experience writing GPU code or FPGA logic, but I have worked (only a little bit) with people who do that type of work.
we're slowly going far afield from topic here :)

No, you're completely right that FPGAs and GPUs are very different beasts. FPGAs are basically high level digital electronics design while GPUs are relatively low-level coding (e.g. C) but still coding. (I am very much oversimplifying here.)

The thing is that I think - think - that in the industry, we're getting to where knowledge of FPGAs is at a relatively static level, but because games/crypto/AI, GPU knowledge is growing. Nvidia GPUs are crazy insane - the V100 offers 100 trillion floating point operations per second (teraflops) across 640 cores ( NVIDIA V100 | NVIDIA ) for example, and the APIs in front of it (e.g. their proprietary CUDA) are pretty well established in AI circles. It's possible to spin up cloud-hosted VMs or even run local GPUs to do development work (at lower performance of course), even for a hobbyist.

In June, 1999, the second largest supercomputer in the world was ASCI Blue Mountain ( ASCI Blue Mountain - Wikipedia ), at a little over 3 teraflops. The available-today-for-preorder Xbox One Series X has a 12 teraflop GPU, meaning a $500 game console your kids want to play Fortnite on has about four times the raw power of a room-filling supercomputer from 21 years go. And things are accelerating.

Anyway ultimately the hardware going in is at a fixed level and there's only so much you can get out of it no matter how many OTA updates you do. Here's hoping whatever they are putting in has some headroom. Automakers tend to be super part cost conscious - that's part of why infotainment systems tend to be slow: cheap-ass processors. I'm hoping Ford is being smart here.
 
OP
macchiaz-o

macchiaz-o

Well-Known Member
First Name
Jonathan
Joined
Nov 25, 2019
Messages
3,234
Reaction score
6,537
Location
Valley of the Sun
First Name
Jonathan
Vehicles
human driven
Country flag
  • Thread starter
  • Thread Starter
  • #9
Anyway ultimately the hardware going in is at a fixed level and there's only so much you can get out of it no matter how many OTA updates you do. Here's hoping whatever they are putting in has some headroom. Automakers tend to be super part cost conscious - that's part of why infotainment systems tend to be slow: cheap-ass processors. I'm hoping Ford is being smart here.
Yup, agreed...

Environmental constraints are another reason for the sluggish and old-looking infotainment in cars. An Xbox One doesn't have to operate at -40F or with massive vibrations and shock, dust, and so on. Choosing parts and components that meet the specs often leads to really junky, outdated looking stuff.
 

GoGoGadgetMachE

Well-Known Member
First Name
Michael
Joined
Jan 23, 2020
Messages
2,882
Reaction score
5,823
Location
Ohio
First Name
Michael
Vehicles
2013 Ford Fusion Energi, 2021 Mach-E First Edition (ordered)
Country flag
Yup, agreed...

Environmental constraints are another reason for the sluggish and old-looking infotainment in cars. An Xbox One doesn't have to operate at -40F or with massive vibrations and shock, dust, and so on. Choosing parts and components that meet the specs often leads to really junky, outdated looking stuff.
hey, you don't know where I'm using my Xbox One! Maybe it's in my airplane's nose cone so I can do Flight Simulator 2020 when it's released while I'm flying an actual plane.

#nope
 

Advertisement





 


Advertisement
Top