BRN Discussion Ongoing

TopCat

Regular

Q&A with SiFive SVP Jack Kang: 5 nm chips with SiFive cores likely to be used in cars by 2025-2026

Judy Lin, DIGITIMES Asia, Taipei
Wednesday 21 September 2022
0


Credit: DIGITIMES

SiFive successfully rolled out a portfolio of automotive CPUs based on the RISC-V standard, and is about to create a new ecosystem of chips, tier-1 auto component providers and OEM partners. DIGITIMES had an exclusive interview with SiFive SVP Jack Kang, who shared what triggered the RISC-V leader's entry into cartech chip IP and the new opportunity SiFive's new portfolio is creating for the supply chain.

Q: What makes the new portfolio stand out from the IP currently available on the market?

We've seen tremendous growth for RISC-V in all market segments over the last few years, and it keeps growing. Automotive is one area where there's a lot of interest now. Why? Because the automotive ecosystem is changing, it's evolving. There's a whole new software ecosystem. A lot of customers have what's called "first party software", which means they're in control of their own software stacks. That means it's easy for them to move to new architectures, especially when that architecture bring significant benefits in power performance, ecosystem choice and things of that nature.

In the RISC-V space, SiFive has the most complete portfolio of products for automotive. There are some other options by vendors who have announced certain point products, but in the RISC-V space this is the most complete portfolio of products for automotive, in terms of supporting everything from 32 bit to 64 bit to vector processing, covering all the different areas for the automotive market. The strength of SiFive is that we're able to provide the entire portfolio. There are multiple different cores within a system, and there are multiple different SOCs, which all need different versions of a CPU. Being able to provide that is very critical for large-scale adoption. Then you get the specific benefits offered by the RISC-V ecosystem combined with the benefits offered by SiFive products. From an ecosystem standpoint, by moving your software to RISC-V you now have freedom of choice. You have lots of open-source software developers working on it, and you're going to have different suppliers offering solutions.

The RISC-V ecosystem is going to grow much faster, and you're not dependent on any one vendor, which means you're going to have more diversity of choices. Then you look at the SiFive products themselves. Our products provide significant power and area advantages at the same performance level as other cores. If you compare to other competing cores from suppliers like ARM and others, SiFive typically offers 30 to 40% better power consumption efficiency, plus a smaller area. This enables more cores on a chip, a lower cost for chips, and lower power consumption.

Q: Besides Renesas, which companies have become partners with SiFive in their new products? Do your IP also have to go through years of validation with the auto makers?

On the open-source point, it's very important to understand that RISC-V is an open standard. That open standard is shared, which means lots of people can do things on it, enabled by a community of openness. And with that there's a lot of open-source software developed for RISC-V. There's a significant ecosystem of partners who develop software; high integrity, fault tolerant software for the automotive space. Now, those are not open source, right? Those partners are crucial partners for the day-to-day grind. By supporting RISC-V, then their ecosystem of hardware partners expands to not just SiFive, but others. This is why RISC-V is mutually beneficial, and there'll be large-scale adoption.

For the partners in our ecosystem, Renesas, along with some 15 other companies, announced support for our automotive launch. We have a couple of other very large customers, tier-1 chip companies, but they are not public, so you have to look for those announcements in the weeks and months to come. But we do have customers for all of these products that we are announcing today.

Our IP is very thoroughly validated on the baseline stuff, and then when we get into the automotive products, there are certain SOP/SOD type of requirements and compliance things that are required.

In many cases, either we're providing those, or we're working with our customers closely to ensure that the solutions meet safety and reliability standards. Because if we look at the automotive space, ultimately, it's about the safety and quality requirements for the final end product and for the system level. So it depends on the customer: some of it is deployed at the system level, some of its deployed in the chip level, some of it is deployed in the IP level. You'll see in the release that we talked about our products being capable and suitable for different types of applications.

SiFive has been around for seven years now. Some of our automotive products have their roots in the success of other embedded products shipping with our IP. It's hard for me to quantify how many years of verification we do for each one, but as IP providers, it is very important that we build and develop trust with our customers.

Q: What inspired SiFive to make entry to the Cartech space? How many years of R&D have you spent in it?

Actually, it was our customers that kept asking us. A couple years ago, we were a little bit hesitant. We were working in other areas, and they kept asking. There's clearly a demand here and a need for something.

Cars have been going through this very rapid technology change. With the electrification of cars and cars becoming a "data center on wheels," cars are becoming very advanced technically. But if you look at semiconductors and the CPU IP available for the space, it has not made the same progress as CPUs and chips for AI or data center chips or mobile chips. The automotive space was a little bit behind, so there was an opportunity for innovation to come in.

Our customers kept pushing until finally we decided we had to take on this market opportunity. And the more we looked into the automotive space, the more we discovered that it is a very good fit for our products and our roadmap, not just the three new solutions that we announced. And we've built a team of automotive experts to help drive this forward.

Q: Which process nodes best suit E6-A, S7-A and X280-A series of processors?

It's pretty broad because we have customers in every automotive node you can think of because RISC-V is very flexible. Now, as you get to higher and higher performing cores, and you start talking about more advanced capabilities, such as ADAS, and you get to L-3, L-4 and beyond, those chips are going into more and more advanced processes because you still need that processing capability.

That creates some additional challenges for functional safety and fault tolerance and has to be handled in other ways. For some of the other functions, maybe they will stay at some of the older nodes. As you get to the higher and higher performance ones and the more advanced nodes, you're going to be pushing into higher and higher performance cores. Some of that will probably be the cores on our roadmap that we haven't announced yet, that will be pushing towards 5 nm or lower nodes because you need that kind of compute.

Q: How long would it take for that core to go into five or below 5 nm?

The auto industry is speeding up, but it still has a longer cycle than consumer or mobile or even the datacenter. For the discussions that we're having now, our IP will be used for chips in 2024 and 2025, which means they won't make it into cars until the model years 2025 or 2026. Right now, they're all trying to accelerate that schedule. That's probably the fastest timeframe that you'd be looking at. It's not like mobile, where you announce something and you see it right away.

Q: And because cybersecurity is the issue that many people concerned about the connected vehicles, how secure is the RISC-V architecture auto processors?

Security is a system issue. One of the very things about RISC-V is from an architecture standpoint, it was developed at a time when cyber security is very important. RISC-V has a clean slate, no legacy of stuff that you're trying to attach onto it. For example, Spectre and Meltdown happened a couple of years ago, but no RISC-V cores were affected by that.

SiFive offers a security model called WorldGuard. This is an example of a security implementation our customers can choose. And there is an advantage to being built on RISC-V also. RISC-V is an open standard, that means its architecture is available for everybody to see. You get the benefit of all of the companies looking at it, and making sure there are no issues and holes, and contributing to the shared standard. At the end, you still have to do the right thing, not just in the CPU, IP, but also in the SOC and the system. Everybody has to pay attention to it. The good news is that security requirements are down to the core, and people have a much better understanding compared to even five years ago or 10 years ago. That's a big advantage that SiFive has: a clean slate for designs.

Q: Intel has this collaboration with you and also with the RISC-V community. The foundry service for all those chips will go to Intel? Or other foundry makers also have a chance.

Intel has certainly been a big proponent for RISC-V, which is great for the ecosystem. They see that this is where a lot of new chip designs are going to happen, and RISC-V will drive value for their foundry. You're going to see this across the board, all foundries are going to benefit from more RISC-V designs. This is where the foundries get to differentiate on their capabilities and features. We'll see how they compete, but it's clear that RISC-V is going to create more opportunities for all the foundries.
I like this bit…..

We have a couple of other very large customers, tier-1 chip companies, but they are not public, so you have to look for those announcements in the weeks and months to come.
 
  • Like
  • Fire
  • Love
Reactions: 29 users

stockduck

Regular
Some interesting components to read maybe?


".....
  • The Arm® Cortex®-M core-based MCX portfolio includes the high-performance MCX N series, cost-optimized and analog-focused MCX A series, low-power wireless connectivity MCX W series and ultra-low power MCX L series
  • The new portfolio features the first instantiation of a specialized, NXP-designed machine learning accelerator to enable high performance inference at the edge ......"
".....The portfolio also features the first instantiation of NXP’s new, specialized neural processing unit (NPU) for accelerating inference at the edge, delivering up to 30x faster machine learning throughput compared to a CPU core alone.
..."

".....
The four series in the MCX portfolio are designed for ease of use, to simplify migration and to scale up or down as needed with maximal software reuse to minimize development cost. The portfolio is based on high-performance Arm Cortex-M cores and integrates a comprehensive set of peripherals for design flexibility. MCX devices feature up to 4 MB of on-chip flash memory, low power cache and advanced memory management controllers, plus up to 1MB of on-chip SRAM to further enhance real-time performance of edge applications.

Machine learning and run-time inference will be supported by NXP’s eIQ® ML software development environment. Developers can utilize the easy-to-use tools offered by eIQ to train ML models targeting either the NPU or the CPU core and deploy them on the MCU. MCX families that are built following NXP’s security-by-design approach will offer secure boot with an immutable root-of-trust, hardware accelerated cryptography and, on select families, a built-in EdgeLock® secure subsystem.

..."

Any Idea?


Also some new development:


"......Expected to be launched first in China from 2023, ProAI will be installed as a supercomputer in heavy-duty vehicles in China and the US to support autonomous driving.
...."


"...... The software supports across 14+ hardware platforms by Qualcomm, Texas Instruments, Renesas and other chipset makers and provides the network to run deep learning-based perception software on V3H, TDA4VM, SA8155 and CV22.
....."
 
  • Like
  • Fire
Reactions: 21 users

Dozzaman1977

Regular
The Brainchip/Moschip joint demonstartion from May is still showing up on the first page of MOSCHIPs website.................
1664748169420.png
 
  • Like
  • Fire
  • Love
Reactions: 36 users

TheFunkMachine

seeds have the potential to become trees.
Excuse me for hyperventilating!!!!!! Please tell me I'm not dreaming!!!!!!



View attachment 17821




Alexa Custom Assistant Will Power BMW’s Voice AI​


Eric Hal Schwartz
on September 30, 2022 at 6:00 am

AddThis Sharing Buttons
Share to Twitter
Share to LinkedInShare to FacebookShare to RedditShare to Email
bmw-amazon.png
BMW and Amazon revealed plans to build BMW’s next-generation voice assistant as an Alexa Custom Assistant platform. s annual Devices and Services launch event. A presentation at Amazon’s annual hardware event showcased how the collaboration will speed up the deployment while reducing costs for BMW, yet drivers will be able to interact with the BMW-branded assistant as though it were built in-house.


BMW Alexa​


“The Alexa technology will enable an even more natural dialogue between driver and vehicle, so drivers can stay focused on the road,” BMW senior vice president of development technical operations Stephan Durach said in announcing the new voice AI with Amazon devices and services business senior vice president Dave Limp. “This will bring the digital experience to an entirely new level.”


BMW didn’t explain how this will affect the recent extension of its long-term partnership with automotive voice AI developer Cerence. Cerence only recently shared its plans for a revamped version of its BMW voice assistant, set to arrive in the BMW 7 and i7 models this fall, The exact models for the Alexa-based assistant haven’t been announced yet. BMW and Amazon only said it would be in the next two years or so. The appeal of designing its voice AI as an Alexa Custom Assistant is not hard to see. It’s all of the customization benefits of independence but powered by all of Alexa’s Amazon support and regular updates. Plus, should BMW’s contributions fall short, the AI will turn to Alexa for help.


“Since we launched Alexa Custom Assistant last year, we’ve continued to improve the solution for automakers and developers. We’re particularly excited about some of the natural language and artificial intelligence advancements we’re delivering, including the capability to effortlessly build multilingual voice personas at scale and enabling uninterrupted voice interactions even when customers are in an area without connectivity,” Amazon smart vehicles vice president Anes Hodzic explained. “We’re also making it easier to create custom skills that are context-aware and proactive.”


BMW has been active in augmenting its voice AI platform with new features like Cerence’s driver-customized wake word. The Cerence deal hasn’t been exclusive, so different models may run on one or more voice AI systems. BMW made Google Assistant accessible in 2017, a year before Alexa became an option. Most recently, BMW began working on incorporating Google’s Android Automotive platform into at least some of its vehicles. It may just be that some of BMWs planned vehicles work best with Amazon’s approach to the technology.


“This cooperation with BMW is a great example of what Alexa Custom Assistant was designed for — to make it faster and easier for companies to develop custom intelligent assistants for virtually any device, without the cost and complexity of building from the ground up,” Limp said.



I think I have missed some connections here, do you mind explaining to me why this Alexa/BMW deal excites you? Is it the fact it can do processing without internet connection hence maybe it’s Akida enabled? Get me excited Bravo! I want in on your hyperventilating. Lol 😂
 
  • Haha
  • Like
Reactions: 13 users
Just a bit of Sunday night fun.

Has anyone noted this on the Brainchip site? The use of the word 'antidote'

www.merriam-webster.com/dictionary/antidote

antidote​

noun

an·ti·dote | \ ˈan-ti-ˌdōt \

Definition of antidote


1: a remedy to counteract the effects of poisonneeded the antidote for the snake's venom
2: something that relieves, prevents, or counteracts

View attachment 17872

So basically Akida is a antidote to counteract computer energy usage, heat and latency.

Once again, Akida is here as a antidote to save the world.🥳🥳🥳

It's safe to say Akida is an antidote and not a sauce. 😆😆😆

It's great to be a shareholder 🏖
Pity they did not develop an antidope for seeing off WANCA’s. I believe that would sell out faster than new iPhones did ten years ago.

My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Haha
  • Like
Reactions: 12 users
I think I have missed some connections here, do you mind explaining to me why this Alexa/BMW deal excites you? Is it the fact it can do processing without internet connection hence maybe it’s Akida enabled? Get me excited Bravo! I want in on your hyperventilating. Lol 😂
This is one of those lots of dots and wishful thinking type connections. The fact that Valeo is supplying ultrasonic sensors to BMW which it is likely have AKIDA onboard gets corporate cheeks rubbing up against each other hopefully in a romantic not uncomfortable fashion.

My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Haha
  • Like
Reactions: 20 users
This is one of those lots of dots and wishful thinking type connections. The fact that Valeo is supplying ultrasonic sensors to BMW which it is likely have AKIDA onboard gets corporate cheeks rubbing up against each other hopefully in a romantic not uncomfortable fashion.

My opinion only DYOR
FF

AKIDA BALLISTA

@uiux's laptop after opening dot joining posts

laptop GIF
 
  • Haha
  • Like
  • Fire
Reactions: 29 users

Jumpchooks

Regular
  • Like
  • Haha
  • Fire
Reactions: 22 users

Jumpchooks

Regular
Hi Brainchip Sales team,
I sent this today to:
Hi Altium Tech Team,
I am private Investor with holdings in BrainChip (ASX:BRN)
www.brainchipinc.com
I'm sure you have heard of them, if not it looks like you guys should talk
 
  • Like
Reactions: 11 users

jk6199

Regular
I can't believe the share price still!

I'm checking everywhere at the moment for extra cash and leaving the calculator open trying to work out if I can afford more, even a small chunk?

There's nothing wrong with spaghetti on toast for a few meals?
 
  • Like
  • Haha
  • Love
Reactions: 21 users

HopalongPetrovski

I'm Spartacus!
  • Haha
  • Like
Reactions: 5 users

Slade

Top 20
Are we expecting a new Brn podcast soon?
 
  • Like
  • Thinking
Reactions: 4 users

Shadow59

Regular
I can't believe the share price still!

I'm checking everywhere at the moment for extra cash and leaving the calculator open trying to work out if I can afford more, even a small chunk?

There's nothing wrong with spaghetti on toast for a few meals?
I'm so with you on this. Just waiting for someone to pay an invoice.
 
  • Like
  • Haha
Reactions: 7 users

HopalongPetrovski

I'm Spartacus!
I can't believe the share price still!

I'm checking everywhere at the moment for extra cash and leaving the calculator open trying to work out if I can afford more, even a small chunk?

There's nothing wrong with spaghetti on toast for a few meals?
IMO. Small time manipulators having a little push and play on a quiet low volume day on the schmarket.
Know they won't get much but a nice setup if further pain in America tonight.
They are hoping to welcome us to their nightmare. 🤣

 
  • Like
  • Haha
  • Fire
Reactions: 7 users
This is one of those lots of dots and wishful thinking type connections. The fact that Valeo is supplying ultrasonic sensors to BMW which it is likely have AKIDA onboard gets corporate cheeks rubbing up against each other hopefully in a romantic not uncomfortable fashion.

My opinion only DYOR
FF

AKIDA BALLISTA
BMW have already partnered with another Neuromorphic AI provider SynSense earlier this year. So I think anything at this stage regarding BMW is wishful thinking.

 
  • Thinking
  • Like
Reactions: 9 users

Townyj

Ermahgerd
BMW have already partnered with another Neuromorphic AI provider SynSense earlier this year. So I think anything at this stage regarding BMW is wishful thinking.

Synsense.jpg


Well who really knows... Hmmm
 
  • Like
  • Fire
  • Love
Reactions: 31 users

Diogenese

Top 20
BMW have already partnered with another Neuromorphic AI provider SynSense earlier this year. So I think anything at this stage regarding BMW is wishful thinking.


The SynSense/BMW arrangement is exploratory, ie, BMW are testing SynSense technology.

"This exploration with BMW in the area of neuromorphic technology will focus on SynSense’s dynamic visual intelligence SoC––Speck, which combines in one single chip SynSense’s low-power SNN vision processor with an event-based sensor. It performs both neuromorphic sensing and computing , and utilizes full asynchronous circuit design. It can be used to capture real-time visual information, recognize and detect objects, and perform other vision-based detection and interaction functions."

So that leaves the door open for better solutions.

Having Swiss origins, SynSense started out in 2017 with analog SNNs and were still going down that path in 2021:

CN113255905A Signal processing method for neurons in spiking neural network and network training method

1664760951000.png




The invention discloses a signal processing method for neurons in a spiking neural network and a network training method. Different from a single-pulse mechanism which is commonly used at present, the multi-pulse mechanism is designed. The neuron signal processing method comprises the following steps: a receiving step: at least one neuron receives at least one path of input pulse sequence; an accumulation step: performing weighted summation based on the at least one path of input pulse sequence to obtain a membrane voltage; and an activation step: when the membrane voltage exceeds a threshold value, determining the amplitude of the pulse excited by the neuron based on the ratio of the membrane voltage to the threshold value. In order to solve the problems of time consumption and low efficiency of a training algorithm caused by an increasing configuration parameter scale, efficient training of the spiking neural network is realized through technical means such as a multi-pulse mechanism, a periodic exponential function proxy gradient and addition of a neuron active inhibition degree as loss, and low power consumption of simulated hardware can still be maintained; and the technical effects of improving the precision and the convergence speed and the like are also brought.

I don't know if SynSense have solutions for the manufacturing variability which plagues analog SNNs.
 
  • Like
  • Love
  • Fire
Reactions: 51 users
BMW have already partnered with another Neuromorphic AI provider SynSense earlier this year. So I think anything at this stage regarding BMW is wishful thinking.

I don’t want to appear to be saying there is an agreement but this is illustrative of how things are not always what they might appear:


Then in what might have been considered impossible Prophesee announces a partnership with Brainchip to “better serve” their customers needs.

As I pointed out before writing something off you need to read all the announcements around BMW and SynSense and it is clear they are not locked in but are exploring possibilities.

I think personally it is wishful thinking to believe that an inferior technology has grasped the opportunity away from Brainchip as many did when Prophesee made its SynSense announcement last year.

My opinion only DYOR
FF

AKIDA BALLISTA
 
Last edited:
  • Like
  • Fire
  • Love
Reactions: 50 users

Diogenese

Top 20
View attachment 17904

Well who really knows... Hmmm
The inclusion of Prophesee is particularly interesting, as SynSense talk of their Speck "which combines in one single chip SynSense’s low-power SNN vision processor with an event-based sensor."

Prophesee has the only DVS/event camera in production, so it seems that SynSense is combining its SNN with a Prophesee event camera.

Even if this were to be successful, there is still radar and LiDaR.

"This exploration with BMW in the area of neuromorphic technology will focus on SynSense’s dynamic visual intelligence SoC––Speck"
 
Last edited:
  • Like
  • Fire
  • Love
Reactions: 26 users

Iseki

Regular
View attachment 17904

Well who really knows... Hmmm
Strange. To my untrained abilities it looks like Synsense has only 2 as yet not granted patent applications, and that after forming an alliance with Prophesee late last year, Prophesee has much more recently formed an alliance with BRN.
Perhaps Prophesee weren't getting what they wanted from Synsense, or perhaps Synsense weren't licensed?

Confirmation greatly appreciated.
 
  • Like
  • Love
Reactions: 10 users
Top Bottom