BRN Discussion Ongoing

[Bravo]

Mercedes is still listed as customer… question is if they want to improve their own AI with this move or we are in the game… time will tell

I personally think Akida 2 fits the Mercedes + Samsung + LG direction exceptionally well.

If Mercedes is shifting toward a heterogeneous compute architecture for next-gen AI-defined vehicles (which the article strongly suggests) then they’ll need both heavyweight compute for generative models and an ultra-low-power layer for continuous sensing. That second layer is exactly where Akida excels.

In-cabin AI will be multimodal and partly generative, combining voice, gaze, gesture, sound events and cabin behaviour. A single large SoC running all of that 24/7 isn’t practical because it would drain power, create heat and require the system to stay constantly awake.

With Akida 2 + TENNs now being positioned as “edge GenAI,” the fit becomes even clearer IMO. Future cabins will rely on small, efficient SLMs that can run locally for privacy, latency and energy reasons. Natural voice interaction, context-aware assistance or conversational feedback from a Samsung or LG in-cabin system won’t require huge LLMs, they'd rely on compact generative or state-space models, which is precisely what Akida 2 is built for.

So while Samsung and LG will handle the heavy compute and platform electronics, Akida could take on a unique, self-learning, sub-watt always-on intelligence that can watch, listen and adapt without waking the main processor. That’s ideal for wake-word detection, anomaly detection, driver and cabin monitoring, and personalised behaviour.

If Mercedes does move to a multi-vendor AI stack, there’s potential for BrainChip’s neuromorphic tech to become the specialised piece that ties the whole system together. At least, this is what I'm hoping.

IMO. DYOR.

 

[smoothsailing18]

I personally think Akida 2 fits the Mercedes + Samsung + LG direction exceptionally well.

If Mercedes is shifting toward a heterogeneous compute architecture for next-gen AI-defined vehicles (which the article strongly suggests) then they’ll need both heavyweight compute for generative models and an ultra-low-power layer for continuous sensing. That second layer is exactly where Akida excels.

In-cabin AI will be multimodal and partly generative, combining voice, gaze, gesture, sound events and cabin behaviour. A single large SoC running all of that 24/7 isn’t practical because it would drain power, create heat and require the system to stay constantly awake.

With Akida 2 + TENNs now being positioned as “edge GenAI,” the fit becomes even clearer IMO. Future cabins will rely on small, efficient SLMs that can run locally for privacy, latency and energy reasons. Natural voice interaction, context-aware assistance or conversational feedback from a Samsung or LG in-cabin system won’t require huge LLMs, they'd rely on compact generative or state-space models, which is precisely what Akida 2 is built for.

So while Samsung and LG will handle the heavy compute and platform electronics, Akida could take on a unique, self-learning, sub-watt always-on intelligence that can watch, listen and adapt without waking the main processor. That’s ideal for wake-word detection, anomaly detection, driver and cabin monitoring, and personalised behaviour.

If Mercedes does move to a multi-vendor AI stack, there’s potential for BrainChip’s neuromorphic tech to become the specialised piece that ties the whole system together. At least, this is what I'm hoping.

IMO. DYOR.

Here is a comment from Linkedin regarding heterogeneous and brainchip

 

[7für7]

Here is an tool for future Mercedes customers
You can translate it in English

KI-Projekt „Ask Mercedes”

www.adesso.de

 

[7für7]

I personally think Akida 2 fits the Mercedes + Samsung + LG direction exceptionally well.

If Mercedes is shifting toward a heterogeneous compute architecture for next-gen AI-defined vehicles (which the article strongly suggests) then they’ll need both heavyweight compute for generative models and an ultra-low-power layer for continuous sensing. That second layer is exactly where Akida excels.

In-cabin AI will be multimodal and partly generative, combining voice, gaze, gesture, sound events and cabin behaviour. A single large SoC running all of that 24/7 isn’t practical because it would drain power, create heat and require the system to stay constantly awake.

With Akida 2 + TENNs now being positioned as “edge GenAI,” the fit becomes even clearer IMO. Future cabins will rely on small, efficient SLMs that can run locally for privacy, latency and energy reasons. Natural voice interaction, context-aware assistance or conversational feedback from a Samsung or LG in-cabin system won’t require huge LLMs, they'd rely on compact generative or state-space models, which is precisely what Akida 2 is built for.

So while Samsung and LG will handle the heavy compute and platform electronics, Akida could take on a unique, self-learning, sub-watt always-on intelligence that can watch, listen and adapt without waking the main processor. That’s ideal for wake-word detection, anomaly detection, driver and cabin monitoring, and personalised behaviour.

If Mercedes does move to a multi-vendor AI stack, there’s potential for BrainChip’s neuromorphic tech to become the specialised piece that ties the whole system together. At least, this is what I'm hoping.

IMO. DYOR.

It will be also interesting what they will bring out here

https://group.mercedes-benz.com/innovations/digitalisation/industry-4-0/mbdfc-humanoid-robots.html

And this.. in which Mercedes has invested and is stakeholder

AI Edge Computing Platforms - AI Edge Computing Hardware

Blaize's edge AI computer platforms bring a new class of efficient, low power, low latency processing to edge computing hardware deployments. Learn more here!

www.blaize.com

 

[Bravo]

It will be also interesting what they will bring out here

https://group.mercedes-benz.com/innovations/digitalisation/industry-4-0/mbdfc-humanoid-robots.html

And this.. in which Mercedes has invested and is stakeholder

AI Edge Computing Platforms - AI Edge Computing Hardware

Blaize's edge AI computer platforms bring a new class of efficient, low power, low latency processing to edge computing hardware deployments. Learn more here!

www.blaize.com

Blaize is being evaluated by Mercedes for higher-end ADAS L4-type platforms, with an automotive-grade chip not expected in production before 2028 (see below excerpt from EETimes).

With Blaize there doesn't seem to be any sign of on-chip learning, whereas BrainChip explicitly markets on-device learning/continual learning as a core feature.

I suppose Mercedes is hedging across different layers of their future compute stack, but if they need something truly self-learning and event- driven, then BrainChip is looking good IMO.



Extract

https://www.eetimes.com/first-ai-chip-startup-goes-public-as-blaize-merges-with-spac/?utm_source=chatgpt.com

 

[7für7]

Blaize is being evaluated by Mercedes for higher-end ADAS L4-type platforms, with an automotive-grade chip not expected in production before 2028 (see below excerpt from EETimes).

With Blaize there doesn't seem to be any sign of on-chip learning, whereas BrainChip explicitly markets on-device learning/continual learning as a core feature.

I suppose Mercedes is hedging across different layers of their future compute stack, but if they need something truly self-learning and event- driven, then BrainChip is looking good IMO.



Extract
View attachment 93048

https://www.eetimes.com/first-ai-chip-startup-goes-public-as-blaize-merges-with-spac/?utm_source=chatgpt.com

Finger-crossed they realise that they need us .. no matter what they want to reach in this segment…! And let’s hope they don’t pull out some rabbit out of the hat

 

[Pom down under]

Finger-crossed they realise that they need us .. no matter what they want to reach in this segment…! And let’s hope they don’t pull out some rabbit out of the hat

 

[7für7]

Would be nice if they could see a usecase to optimise their processes via Akida also in this field… we know Akida is fantastic in objekt recognition and we could also help in improving manufacturing Prozesses and so on… thoughts on that?

Mercedes-Benz beschleunigt KI-Transformation mit Celonis

Mercedes-Benz beschleunigt KI-Transformation mit Celonis

www.celonis.com

English

https://www.businesswire.com/news/home/20251104923787/en/Mercedes-Benz-Accelerates-AI-Driven-Transformation-With-Celonis

 

[7für7]

It’s old but interesting though


Event-based Vision Sensors (“EVS”) are said to have been modeled after human optic nerves. Completely different from previous frame-based image sensors, EVS can perceive only changes that have occurred in subjects, in a form like a path or trajectory, so they do not use the concept of “frames” in which sensing targets are captured as images, and have the characteristic of being able to monitor even objects moving at high speed in real-time. Also, in keeping with “their origin of being modeled after optic nerves, EVS are anticipated to function as the eyes of AI or robots, and it is believed that if they can be commercialized, they will allow robots to make judgments and decisions even more quickly. These are the EVS that have been developed by the Sony Semiconductor Solutions Group (“SSStheour Group”) in collaboration with PROPHESEE, which are expected to breathe new life into various fields including the industrial machinery sector. Their development involved difficulties and discoveries that were only possible precisely because they are entirely new types of vision sensors.”

Event-based Vision Sensors that can function like human optic nerves and monitor only changes in subjects will open up vast new possibilities for AI and robots | Feature｜Sony Semiconductor Solutions Group

Introducing special contents of Sony Semiconductor Solutions Group.

www.sony-semicon.com

 

[manny100]

For auto safety features focused on driver monitoring and adaptive sensing, Akida with on-chip learning is far more suitable than Blaize.
For other features Blaize may be more suitable.
It may be that Mercedes is not ready yet to let AKIDA safety features on the road and is still testing it.
I would not in any way expect that AKIDA will be the only chip in the car. It's likely it will be horses for courses.
I think ' adaptable on chip learning' would be compulsory for auto safety.

 

[7für7]

For auto safety features focused on driver monitoring and adaptive sensing, Akida with on-chip learning is far more suitable than Blaize.
For other features Blaize may be more suitable.
It may be that Mercedes is not ready yet to let AKIDA safety features on the road and is still testing it.
I would not in any way expect that AKIDA will be the only chip in the car. It's likely it will be horses for courses.
I think ' adaptable on chip learning' would be compulsory for auto safety.

It would already be enough if they implemented Akida just to reduce energy-hungry systems so EVs and plug-ins can achieve a greater range. We can support many different use cases – it depends on how each customer wants to use Akida. We doesn’t need to be implemented in security relevant systems like autonomous driving and stuff like this…
For example

Reducing compute load for sensors (cameras, radar, lidar, interior monitoring) I don’t mean to run the system… just optimising the Prozess

Event-based processing instead of continuous full-load processing

“Guardian” function: Akida runs lightweight and only wakes up CPU/GPU when needed

Less or more efficient “cloud”/data communication (more on-device AI)

Lower overall power consumption of ECUs and SoCs

Less heat generation…less cooling effort for electronics

More efficient energy management in the vehicle (driving profile, recuperation, HVAC)

Ability to use smaller/more efficient power and board-net components

Overall reduction of continuous electronic power draw… more range for EVs

I think if the automotive sector would use us only in those points, we would win a lot of trust at the industry