BRN Discussion Ongoing

[Dhm]

Brainchip / Prophesee podcast
From 16.40 minute mark.
For past 3 years working on combining regular frame based with event based sensors in smart phones to solve motion blur.
Rob Telson confirming soon after that is where the Brainchip / Prophesee collaboration is headed.

At the 33:03 mark the bloke states the frame rate is up to 10,000 per second. Can we do this??

Edit: I clicked on the wrong link, sorry. I meant to highlight the Qualcomm video.

 

[SharesForBrekky]

Must be missing something but I'm trying to figure out what the role is a hybrid of?

I wondered if was just the role has a couple of area crossovers outside just project management or if refers to the actual product / projects being a hybrid AI?

View attachment 27490

Hey FMF, I think it may just mean hybrid working:

"In a hybrid work model, employees have more flexibility to get work done when, how, and where they're most productive. This means flexible schedules and locations. For example, some people work best early in the morning while others do better in the evening."

 

[Fullmoonfever]

Hey FMF, I think it may just mean hybrid working:

"In a hybrid work model, employees have more flexibility to get work done when, how, and where they're most productive. This means flexible schedules and locations. For example, some people work best early in the morning while others do better in the evening."

Cheers fair explanation.

I took it possible other ways obviously.

 

[Stable Genius]

At the 33:03 mark the bloke states the frame rate is up to 10,000 per second. Can we do this??

Hi @Dhm

I just listened to the whole podcast with Prophesee again. It is fantastic!

Brainchip and Prophesee are complimentray technologies. A bit like in Show me the money movie: “You complete me!”

Well worth a listen!

 

[Dhm]

Hi @Dhm

I just listened to the whole podcast with Prophesee again. It is fantastic!

Brainchip and Prophesee are complimentray technologies. A bit like in Show me the money movie: “You complete me!”

Well worth a listen!

Hi @Stable Genius I replied to the wrong @Taproot post. I meant to talk about the Qualcomm video, although our own Episode 20 is uplifting and of course dovetails into the whole potential relationship. I was making the point on the Qualcomm video about 10,000fps and if we could do this.

So much great work sleuthing by the 1000 eyes, well done chaps and chapetts!

 

[Bravo]

Hi @Stable Genius I replied to the wrong @Taproot post. I meant to talk about the Qualcomm video, although our own Episode 20 is uplifting and of course dovetails into the whole potential relationship. I was making the point on the Qualcomm video about 10,000fps and if we could do this.

So much great work sleuthing by the 1000 eyes, well done chaps and chapetts!

Hi @Dhm, this slide shows it's measuring events at sub-millisecond time-scale.

It's a bit confusing because at the 36 min mark the presenter talks about the RGB camera operating at 30-60 frames per second and then when it comes to him mentioning what the Prophesee camera can do, it sounds like he is saying it works 10 to100 x faster than the RGB camera, rather than 10,000 times faster???

 

[Taproot]

Hi @Stable Genius I replied to the wrong @Taproot post. I meant to talk about the Qualcomm video, although our own Episode 20 is uplifting and of course dovetails into the whole potential relationship. I was making the point on the Qualcomm video about 10,000fps and if we could do this.

So much great work sleuthing by the 1000 eyes, well done chaps and chapetts!

PROPHESEE | Metavision Technologies

REVEAL THE INVISIBLE with the world's most advanced neuromorphic vision system, inspired by human vision and built on the foundation of neuromorphic engineering. PROPHESEE gives Metavision to machines, revealing what was previously invisible to them.

www.prophesee.ai

 

[AARONASX]

Must be missing something but I'm trying to figure out what the role is a hybrid of?

I wondered if was just the role has a couple of area crossovers outside just project management or if refers to the actual product / projects being a hybrid AI?

View attachment 27490

I think Hybrid is someone that can work remotely (has the option) from the office.

After Covid I'm now a Hybrid where I only go into the office 1-2 days a week, if that!

 

[Townyj]

Hi @Stable Genius I replied to the wrong @Taproot post. I meant to talk about the Qualcomm video, although our own Episode 20 is uplifting and of course dovetails into the whole potential relationship. I was making the point on the Qualcomm video about 10,000fps and if we could do this.

So much great work sleuthing by the 1000 eyes, well done chaps and chapetts!

You need to remember that its Prophesee not Brainchip providing the product. Prophesee can do 10k fps, we might be some of the secret sauce behind the low power and provide some of the heavy lifting for processing etc.

 

[Stable Genius]

Hi @Stable Genius I replied to the wrong @Taproot post. I meant to talk about the Qualcomm video, although our own Episode 20 is uplifting and of course dovetails into the whole potential relationship. I was making the point on the Qualcomm video about 10,000fps and if we could do this.

So much great work sleuthing by the 1000 eyes, well done chaps and chapetts!

Hi @Dhm

I think the whole point of Prophesee was that it doesn’t operate at FPS. I’m not the most qualified to speak of this but FPS is the old/usual method. Like when you could draw little action pictures on the bottom corner of a book and then flip the pages to create a moving pictures. The faster you would flip the pages increased the FPS. Thats what we did as kids; it was fun!

Prophesee has the shutter constantly open, which allows more light in, whereas to capture fast images you need a faster shutter speed but then less light enters the aperture and it is a darker image. It only acknowledges changes in the image. I think that’s why it’s 1000fps equivalent: because the frame or opening and closing of the shutter doesn’t actually occur.

Brainchip complements that process by also only acknowledging the changes in the image.

Feel free to correct me if I’m wrong with my understanding of this.

Cheers!

 

[Diogenese]

Must be missing something but I'm trying to figure out what the role is a hybrid of?

I wondered if was just the role has a couple of area crossovers outside just project management or if refers to the actual product / projects being a hybrid AI?

View attachment 27490

Hi Fmf,

I've seen the term "hybrid" in the context of AI used to refer to a mixture of analog (MemRistor [WBT?]) and digital NN, usually with at least the output layer being digital.

Analog is theoretically closer to the way real synapses work than digital, but has problems with repeatability in the manufacturing process.

Were there any more details on what the job entails?

 

[Damo4]

At the 33:03 mark the bloke states the frame rate is up to 10,000 per second. Can we do this??

Edit: I clicked on the wrong link, sorry. I meant to highlight the Qualcomm video.

Short answer is YES mate, it is exactly what Akida + Event based cameras can do.
However it is not FPS so much as the ability to recognise pixel changes at high speed.
I believe they are using the term FPS to make it more understandable.


Edit: @Stable Genius Explains it perfectly above, sorry for the duplication

 

[skutza]

 

[Boab]

Hi @Dhm

I think the whole point of Prophesee was that it doesn’t operate at FPS. I’m not the most qualified to speak of this but FPS is the old/usual method. Like when you could draw little action pictures on the bottom corner of a book and then flip the pages to create a moving pictures. The faster you would flip the pages increased the FPS. Thats what we did as kids; it was fun!

Prophesee has the shutter constantly open, which allows more light in, whereas to capture fast images you need a faster shutter speed but then less light enters the aperture and it is a darker image. It only acknowledges changes in the image. I think that’s why it’s 1000fps equivalent: because the frame or opening and closing of the shutter doesn’t actually occur.

Brainchip complements that process by also only acknowledging the changes in the image.

Feel free to correct me if I’m wrong with my understanding of this.

Cheers!

The Qualcomm guy says "these types of cameras don't work off frame rates" or words to that effect.

 

[kayaker]

Hi Fmf,

I've seen the term "hybrid" in the context of AI used to refer to a mixture of analog (MemRistor [WBT?]) and digital NN, usually with at least the output layer being digital.

Analog is theoretically closer to the way real synapses work than digital, but has problems with repeatability in the manufacturing process.

Were there any more details on what the job entails?

If I am correct below is an example of "hybrid SNN

3.3. Tianjic​

The Tianjic project (2019, Tsinghua University) (Pei et al., 2019) is the first hybrid chip that can work effectively with both ANNs and SNNs. This possibility comes from the reuse of the same parts of circuits to work with different types of neural networks. The additional overhead for such versatility is only 3% of the chip area. Thus, using the Tianjic chip, it is possible to combine architectures of neural networks of different nature (ANN and SNN) within one system. One Tianjic chip contains 156 neural cores, simulating 40,000 neurons and 10,000,000 synapses. Each core contains 22 Kbyte of SRAM. The digital data bus is used for communication between the cores, and the signals are represented as AER packets. Scaling is achieved by combining chips into a 2D mesh network. On-chip learning is not supported. The neural network must be pre-trained on another platform (most frequently, GPU) and translated into the Tianjic configuration to work in the inference mode. Running SNN on Tianjic is 22 times faster and 10,000 times more energy efficient than on GPU. For ANNs, the gains are also significant:

• LSTM networks are 467 times more energy efficient,
• MLPs are 723 times more energy efficient and 35 times faster in terms of frame rates,
• CNNs are 53 times more energy efficient and 101 times faster in terms of frame rates.

An example of using Tianjic chip to create a bicycle motion control system is presented in Pei et al. (2019). This system, implemented on only one Tianjic chip, includes real-time object detection (CNN), object tracking (CANN), voice control (SNN), obstacle avoidance, and balance control (MLP). Another SNN, called a Neural State Machine (NSM), was used to integrate neural networks with each other.

 

[Fullmoonfever]

Hi Fmf,

I've seen the term "hybrid" in the context of AI used to refer to a mixture of analog (MemRistor [WBT?]) and digital NN, usually with at least the output layer being digital.

Analog is theoretically closer to the way real synapses work than digital, but has problems with repeatability in the manufacturing process.

Were there any more details on what the job entails?

Hey D

Yeah, seen the same term used that's why pondered the use of the word.

Other posters have fair explanation but the role details as below fwiw.

Project Manager - Hybrid​

BrainChip Laguna Hills, CA
7 hours ago 29 applicants​

Home

Welcome to Brainchip, where AI lives at the edge!

brainchipinc.com

Job Title: Project Manager
Reports to: Chief Marketing Officer

**No Agencies Please**

SUMMARY
The Project Manager is responsible for planning and overseeing projects from the initial ideation through to completion.This position will coordinate people and processes to deliver projects on time, within budget, and with the desired outcomes aligned to objectives. This role reports to the Chief Marketing Officer.

ESSENTIAL JOB DUTIES AND RESPONSIBILITIES:
Immediate (for 2.0 launch)

• Coordinating with cross-discipline team members to make sure that all parties are on track with project requirements, deadlines, and schedules.
• Meeting with project team members to identify and resolve issues.
• Submitting project deliverables and ensuring that they adhere to quality standards.
• Preparing status reports by gathering, analyzing, and summarizing relevant information.
• Establishing effective project communication plans and ensuring their execution.
• Facilitating change requests to ensure that all parties are informed of the impacts on schedule and budget.
• Coordinating the development of collateral – such as user manuals, training materials and launch materials
• In the future, this may involve other documents as needed to enable successful implementation and turnover of the process or system to the clients.

For future activities (as we deliver product)
· Obtaining customer acceptance of project deliverables.
· Managing customer satisfaction within the project transition period.
· Conducting post-project evaluation and identifying successful and unsuccessful project elements.
· Identifying and developing new opportunities with clients
· ERP project oversight.

Education/Experience/Qualifications:
• BS/MS in a related field.
• 10+ years of experience in digital logic design
• Project Management Professional (PMP) certification is a plus.
• Over 10 years of proven experience in project management.
• Ability to lead project teams of various sizes and see them through to completion.
• Strong understanding of formal project management methodologies.
• Experience as a software/h/w product project manager or launch project manager.
• Able to complete projects in a timely manner.
• Budget management experience.

 

[Diogenese]

PROPHESEE | Metavision Technologies

REVEAL THE INVISIBLE with the world's most advanced neuromorphic vision system, inspired by human vision and built on the foundation of neuromorphic engineering. PROPHESEE gives Metavision to machines, revealing what was previously invisible to them.

www.prophesee.ai

Hi @Dhm, this slide shows it's measuring events at sub-millisecond time-scale.

It's a bit confusing because at the 36 min mark the presenter talks about the RGB camera operating at 30-60 frames per second and then when it comes to him mentioning what the Prophesee camera can do, it sounds like he is saying it works 10 to100 x faster than the RGB camera, rather than 10,000 times faster???

View attachment 27493

10,000 fps equivalent. I take this to mean that Prophesee's DVS event camera can capture movement (change in light impinging on a pixel) detected by individual pixels at 10,000 Hz, not a full screen of pixels in a frame. DVS cameras do not have a shutter, so the photoreceptor plate is continuously exposed to the field of view. The thing which would limit the speed which a DVS could capture movement would be the response time of the pixels unencumbered by any frame rate - without the inherent delay of the fixed frame configuration of normal video. The pixels fire asynchronously as the light impinging on the individual pixels changes.

Going back to Prophesee's comments about Akida, Akida can accept asynchronous input from individual pixels. It does not need to wait for a full frame of image data. It is able to receive individual "events" as they occur. So it seems from Prophesee's comments, Akida is capable of matching the performance of the Prophesee DVS, something that frame-based system cannot do.

On the other hand, nViso has tested Akida with framed video to over 1.6 k fps.

 

[Diogenese]

If I am correct below is an example of "hybrid SNN

3.3. Tianjic​

The Tianjic project (2019, Tsinghua University) (Pei et al., 2019) is the first hybrid chip that can work effectively with both ANNs and SNNs. This possibility comes from the reuse of the same parts of circuits to work with different types of neural networks. The additional overhead for such versatility is only 3% of the chip area. Thus, using the Tianjic chip, it is possible to combine architectures of neural networks of different nature (ANN and SNN) within one system. One Tianjic chip contains 156 neural cores, simulating 40,000 neurons and 10,000,000 synapses. Each core contains 22 Kbyte of SRAM. The digital data bus is used for communication between the cores, and the signals are represented as AER packets. Scaling is achieved by combining chips into a 2D mesh network. On-chip learning is not supported. The neural network must be pre-trained on another platform (most frequently, GPU) and translated into the Tianjic configuration to work in the inference mode. Running SNN on Tianjic is 22 times faster and 10,000 times more energy efficient than on GPU. For ANNs, the gains are also significant:

• LSTM networks are 467 times more energy efficient,
• MLPs are 723 times more energy efficient and 35 times faster in terms of frame rates,
• CNNs are 53 times more energy efficient and 101 times faster in terms of frame rates.

An example of using Tianjic chip to create a bicycle motion control system is presented in Pei et al. (2019). This system, implemented on only one Tianjic chip, includes real-time object detection (CNN), object tracking (CANN), voice control (SNN), obstacle avoidance, and balance control (MLP). Another SNN, called a Neural State Machine (NSM), was used to integrate neural networks with each other.

Thanks Kayaker,

I think I have bamboozled myself with science. Apparently, the covid flexible working conditions is the more probable answer.

 

[chapman89]

@Diogenese what’s your thoughts on this video regarding snapdragon and after taking into account what was said on the Prophesee Brainchip podcast about Brainchip being the other half of the story they were needing to make it so great?

We know that the Prophesee Brainchip relationship dates back to at least mid 2021 and the Prophesee Qualcomm relationship could pre date that, but it just seems that it is likely containing some akida?

 

[Bravo]

@Diogenese what’s your thoughts on this video regarding snapdragon and after taking into account what was said on the Prophesee Brainchip podcast about Brainchip being the other half of the story they were needing to make it so great?

We know that the Prophesee Brainchip relationship dates back to at least mid 2021 and the Prophesee Qualcomm relationship could pre date that, but it just seems that it is likely containing some akida?

Does anyone know if we can go down to a 3nm process?

Snapdragon 8 Gen 3: Everything we know so far​

Qualcomm Snapdragon 8 Gen 3 news and rumours

By Martyn Casserly
Contributor, Tech Advisor JAN 12, 2023 3:44 PM GMT

Qualcomm may have only released its Snapdragon 8 Gen 2 in November 2022, but development of its successor is already underway.
Here’s all we know so far about the Snapdragon 8 Gen 3.

When will the Snapdragon 8 Gen 3 be released?​

We have no official date so far confirming the arrival of the new chip, but Qualcomm has established a pretty regular release schedule over the past few years, so we think it’s highly likely that the Snapdragon 8 Gen 3 will be released in November or December 2023.
This would be in keeping with all of the Snapdragon 8 mobile processors, which made their debuts around the same time:

• Snapdragon 8 Gen 1 – December 2021
• Snapdragon 8 Gen 2 – November 2022

We haven’t heard any reports that this schedule has been disrupted, so its looks good for a winter release this year.

What new features will we see in the Snapdragon 8 Gen 3?​

We haven’t seen much in the way of specs or any definite enhancements that Qualcomm is working on, but we do know one thing just about for certain and this is that the new processor will be based on a 3nm process.
This has been at the centre of some controversy that has surrounded the early reports around the Gen 3 processor. Samsung was the first to achieve this size, but initially it was reported to have only managed to return a yield of 20% per wafer, which obviously isn’t efficient and would result in increased prices for the processors. Qualcomm has now reportedly moved production to Taiwanese manufacturer TSMC, which has managed a much higher 75-80% return.
Samsung did improve its output though, as GSMArena states that once the Korean giant had partnered with US company Silicon Frontline Technology the yield increased to around 60-70%.

All of this has resulted in various reports saying that production will be either all via TSMC or a combination of both. Only time will tell who get to make the new Snapdragon 8 Gen 3.
So, what will a 3nm process bring that makes it worth all this hassle? Well, GizChina reports that the new process should deliver improved performance, plus an important 35% improvement in energy efficiency, with its obvious benefits on battery life.

As processors are so advanced these days, these two additions should make it all worthwhile, plus the marginally smaller architecture could free up space inside devices for more cooling elements or other benefits. Every little helps.
It also seems to be a given that the chip will include support for the new Snapdragon Satellite tech that Qualcomm announced at CES 2023, allowing emergency communications without normal mobile signal – and eventually even basic two-way messaging. This is actually already included in the 8 Gen 2, and has been made available to manufacturers to include in new handsets, so it seems certain it will appear in the next generation too.

Which phones will use the Snapdragon 8 Gen 3?​

Obviously, with the release date still far in the distance, we don’t have any confirmed reports of which phones will come with the Gen 3 chip onboard, but going by past history we can make a few confident predictions.
With the previous generation, some of the first devices out of the gate were the Vivo X90 Pro+, Xiaomi 13 and Xiaomi 13 Pro, and OnePlus 11. We would be pretty certain that those phones’ successors will also be racing to get the new silicon in their flagships for 2024.
As with most years, the Qualcomm processor should also make it to many other flagships, including from companies like Samsung, Motorola, and Oppo.
To see which phones those newly equipped with the Snapdragon 8 Gen 3 will have to beat, check out our current picks of the best smartphones, along with the best upcoming phones in 2023.

Qualcomm's Snapdragon 8s Gen 3 brings AI to more affordable phones

The 8s Gen 3 isn't quite as powerful as the flagship Snapdragon 8 Gen 3, but it should offer a nice performance boost for mid-range handsets

www.techadvisor.com