BRN Discussion Ongoing

[Diogenese]

we know that we had chips manufactured by TSMC and global foundry before, and we are also joined with Intel Foundry Services to advance edge-based neuromorphic AI

Hi Stan,

The EdgeBox will have its own CPU to perform other tasks including utilizing the output from Akida. as well as configuring the Akida chip, so Akida 1500 could be used. 1500 can be used with practically any CPU. GF made the first lot of 1500s in 22 nm FDSoI, so they will have the tapeout data ready to go if we need more chips.

It seems premature to expect that Akida 2 will be used commercially, so there's a good chance it will be 1500.

This is the VVDN Nvidia EdgeBox:

https://www.vvdntech.com/en-us/vision/jetson-edge-box
Edge AI Box is developed and manufactured by VVDN for building your next-gen autonomous machine at the edge. The box includes custom carrier board by VVDN along with NVIDIA powerful NVIDIA® Jetson Xavier NX™ Module.

Module	Xavier NX (8GB / 16GB production version)
AI Performance	21 TOPS
GPU	NVIDIA Volta architecture with 384 NVIDIA CUDA® cores and 48 Tensor cores
CPU	6-core NVIDIA Carmel ARM®v8.2 64-bit CPU 6 MB L2 + 4 MB L3
DL Accelerator	2x NVDLA Engines
Vision Accelerator	7-Way VLIW Vision Processor
Camera	2x2 Lane MIPI CSI-2 DPHY 1x4 Lane MIPI CSI-2 DPHY
USB	4x USB 3.1, 1x USB 2.0 Micro-B
Network Connectivity	1x M.2 Key E for WIFI/BT (WIFI/BT Module included) 1x Gigabit Ethernet
Storage	16 GB eMMC 5.1 1x M.2 Key M for NVMe

Maybe Akida will substitute for the GPU (384 CUDA cores + 48 tensor cores) and the Deep Learning Accelerator?

 

[jk6199]

I somehow bought more today?

Luckily I have a NDA agreement that my better half, does not yet know this detail.

Hopefully this non disclosure remains for my sake.

Good luck everyone, hopefully this lowering price reverses soon!

 

[Stockbob]

Hi Stan,

The EdgeBox will have its own CPU to perform other tasks including utilizing the output from Akida. as well as configuring the Akida chip, so Akida 1500 could be used. 1500 can be used with practically any CPU. GF made the first lot of 1500s in 22 nm FDSoI, so they will have the tapeout data ready to go if we need more chips.

It seems premature to expect that Akida 2 will be used commercially, so there's a good chance it will be 1500.

This is the VVDN Nvidia EdgeBox:

https://www.vvdntech.com/en-us/vision/jetson-edge-box
Edge AI Box is developed and manufactured by VVDN for building your next-gen autonomous machine at the edge. The box includes custom carrier board by VVDN along with NVIDIA powerful NVIDIA® Jetson Xavier NX™ Module.

Module	Xavier NX (8GB / 16GB production version)
AI Performance	21 TOPS
GPU	NVIDIA Volta architecture with 384 NVIDIA CUDA® cores and 48 Tensor cores
CPU	6-core NVIDIA Carmel ARM®v8.2 64-bit CPU 6 MB L2 + 4 MB L3
DL Accelerator	2x NVDLA Engines
Vision Accelerator	7-Way VLIW Vision Processor
Camera	2x2 Lane MIPI CSI-2 DPHY 1x4 Lane MIPI CSI-2 DPHY
USB	4x USB 3.1, 1x USB 2.0 Micro-B
Network Connectivity	1x M.2 Key E for WIFI/BT (WIFI/BT Module included) 1x Gigabit Ethernet
Storage	16 GB eMMC 5.1 1x M.2 Key M for NVMe

Maybe Akida will substitute for the GPU (384 CUDA cores + 48 tensor cores) and the Deep Learning Accelerator?

Had the same thought Dio, but Tony's reply states Akida 1000 so thought maybe not, would be great if it is 1500 as Sean and comp have said these could be used by partners in cards or modules of thier own, timeline and everything fits nicely with 1500. But hey who cares which generation is used , get the box out there in the real world and bring a product to market.

 

[TECH]

Hi Stan,

Yes. Supplying chips is a faster route to revenue, if more time consuming and requiring more effort while being less profitable. I thought Sean pulled the wrong rein when they dropped chip supply because it excluded a huge number of potential customers, but it seems that the penny has now dropped.

If pre-orders are to be in 2023, we should see some income early in 2024. Let's hope we get a premium on the chips. After all, we will be substituting for Nvidia.

I wonder who will be supplying the model libraries?

The libraries will be determined by the specific applications, and will need to be adapted to Akida's format.

Libraries can be built up by training Akida, or, if existing libraries are suitable, they can be converted to Akida format (and supplemented by on-chip learning if needed).

So wonders will never cease, as mentioned to me, the company was still pursuing potential markets that AKD 1000 could fit in very nicely.

So much for being "too narrow" in it's offering, "a failure"....such comments I personally found insulting to Peter, a life body of work that
passed with flying colours and in fact beyond initial expectations, not forgetting first-run engineering samples conducted by TSMC.

As Sean isn't quite 2 years into his 5 year plan, chip sales embedded into other companies products may well have been the first of
many escape clauses embedded in his delivery to the Board back in late 2021, purely a tongue n cheek comment by the Texta

Sean deciding to visit South Korea himself, is a signature beckoning I wonder, someone must like our IP surely

 

[Boab]

I somehow bought more today?

Luckily I have a NDA agreement that my better half, does not yet know this detail.

Hopefully this non disclosure remains for my sake.

Good luck everyone, hopefully this lowering price reverses soon!

Snap

 

[stan9614]

Hi Stan,

Yes. Supplying chips is a faster route to revenue, if more time consuming and requiring more effort while being less profitable. I thought Sean pulled the wrong rein when they dropped chip supply because it excluded a huge number of potential customers, but it seems that the penny has now dropped.

If pre-orders are to be in 2023, we should see some income early in 2024. Let's hope we get a premium on the chips. After all, we will be substituting for Nvidia.

I wonder who will be supplying the model libraries?

The libraries will be determined by the specific applications, and will need to be adapted to Akida's format.

Libraries can be built up by training Akida, or, if existing libraries are suitable, they can be converted to Akida format (and supplemented by on-chip learning if needed).

not sure about the library part, too technical for me.

regarding to Tony's comment on revenue style with VVDN, although he did not mention in the email, but it seems (just my own speculation) to me any partners who are co-developing products with Brainchip will be buying Akida 1000 or 1500 chips from us.

then I start to think about:
ANT61 is launch their product powered by akida in Jan 2024.
Teksun is working on the next-generation AIoT devices powered by akida
Circle8 Clean Technologies and AVID Group to develop “Smart Bins" powered by akida
GMAC Intelligence is building a drive-through automation solution for quick-service restaurants (QSRs) powered by akida
Lorser Industries said they will use BrainChip’s Akida™ technology to deliver neuromorphic computing solutions for software- defined radio (SDR) devices.

and many more announced development partnership...

I hope the management could put all these partner names into the presentation slides next time, and group them into IP licence customers and Chip sales customers. there is no reason why we cannot show off these names publicly as much as we have already did in news published here and there.

I personally do not believe majority of australian investors understood anymore than 30% of what Brainchip has accomplished in the past two years and why these achievement will lead to revenue surge in the very near future.

 

[Diogenese]

Had the same thought Dio, but Tony's reply states Akida 1000 so thought maybe not, would be great if it is 1500 as Sean and comp have said these could be used by partners in cards or modules of thier own, timeline and everything fits nicely with 1500. But hey who cares which generation is used , get the box out there in the real world and bring a product to market.

Thanks Stockbob,

I missed/forgot/overlooked that. So, mutatis mutandis, it could well be Socionext/TSMC or some friend of VVDN (or ARM, or IFS)?

VVDN do incorporate ARM in their Nvidia EdgeBox.

The suspenders are killing me.

 

[stan9614]

Had the same thought Dio, but Tony's reply states Akida 1000 so thought maybe not, would be great if it is 1500 as Sean and comp have said these could be used by partners in cards or modules of thier own, timeline and everything fits nicely with 1500. But hey who cares which generation is used , get the box out there in the real world and bring a product to market.

Hi Stan,

The EdgeBox will have its own CPU to perform other tasks including utilizing the output from Akida. as well as configuring the Akida chip, so Akida 1500 could be used. 1500 can be used with practically any CPU. GF made the first lot of 1500s in 22 nm FDSoI, so they will have the tapeout data ready to go if we need more chips.

It seems premature to expect that Akida 2 will be used commercially, so there's a good chance it will be 1500.

This is the VVDN Nvidia EdgeBox:

https://www.vvdntech.com/en-us/vision/jetson-edge-box
Edge AI Box is developed and manufactured by VVDN for building your next-gen autonomous machine at the edge. The box includes custom carrier board by VVDN along with NVIDIA powerful NVIDIA® Jetson Xavier NX™ Module.

Module	Xavier NX (8GB / 16GB production version)
AI Performance	21 TOPS
GPU	NVIDIA Volta architecture with 384 NVIDIA CUDA® cores and 48 Tensor cores
CPU	6-core NVIDIA Carmel ARM®v8.2 64-bit CPU 6 MB L2 + 4 MB L3
DL Accelerator	2x NVDLA Engines
Vision Accelerator	7-Way VLIW Vision Processor
Camera	2x2 Lane MIPI CSI-2 DPHY 1x4 Lane MIPI CSI-2 DPHY
USB	4x USB 3.1, 1x USB 2.0 Micro-B
Network Connectivity	1x M.2 Key E for WIFI/BT (WIFI/BT Module included) 1x Gigabit Ethernet
Storage	16 GB eMMC 5.1 1x M.2 Key M for NVMe

Maybe Akida will substitute for the GPU (384 CUDA cores + 48 tensor cores) and the Deep Learning Accelerator?

hi, Dio and stockbob
yes I think given VVDN is almost at the end of its product design cycle, i would assume the amount of time needed to test a chip for its new product design, VVDN would only had enough time to examine on Akida 1000. Hope I am wrong though.

and yes, Tony did mention it is akida1000 for vvdn

 

[Pom down under]

RA8M1 - 480MHz Arm Cortex-M85 Based Microcontroller with Helium and TrustZone

The RA8M1 MCU Group features the industry’s first 32-bit microcontrollers based on the Arm® Cortex®-M85 (CM85) core, delivering over 3000 CoreMark at 480 MHz.

www.renesas.com

 

[Schnitzel lover]

"The Edge box will be available for presale from BrainChip and VVDN later this year."

Real revenue may not show up until Q1 2024 but a promising sign of things to come!

Hopefully on a large scale

 

[Diogenese]

not sure about the library part, too technical for me.

regarding to Tony's comment on revenue style with VVDN, although he did not mention in the email, but it seems (just my own speculation) to me any partners who are co-developing products with Brainchip will be buying Akida 1000 or 1500 chips from us.

then I start to think about:
ANT61 is launch their product powered by akida in Jan 2024.
Teksun is working on the next-generation AIoT devices powered by akida
Circle8 Clean Technologies and AVID Group to develop “Smart Bins" powered by akida
GMAC Intelligence is building a drive-through automation solution for quick-service restaurants (QSRs) powered by akida
Lorser Industries said they will use BrainChip’s Akida™ technology to deliver neuromorphic computing solutions for software- defined radio (SDR) devices.

and many more announced development partnership...

I hope the management could put all these partner names into the presentation slides next time, and group them into IP licence customers and Chip sales customers. there is no reason why we cannot show off these names publicly as much as we have already did in news published here and there.

I personally do not believe majority of australian investors understood anymore than 30% of what Brainchip has accomplished in the past two years and why these achievement will lead to revenue surge in the very near future.

Yes. A consolidated list of published wins with partner/customer/licencee name and product/tech field would be useful in allaying the down rampers. Potential time to market and addressable market size would be the icing on the iceberg. Surely our marketing dept could whip this up off the cuff.

 

[Iseki]

Sad.
They tell us sales traction will start with the release of AKIDA 2.
Clearly it didn't because -
now we're back trying to flog the remaining AKIDA 1000, all 6900 of the ones we have left -
in a box we have to co-develop with VVDN.

When Sean can on board he told of the story of Betamax vs VHS. And we have come out as Betamax - a great technology but everyone else is signing up with NVIDIA.

The board's pride is causing our fall.

Here's an idea - let's find someone to co-develop the Akida2000 chip.

 

[Justchilln]

Hi Stan,

Yes. Supplying chips is a faster route to revenue, if more time consuming and requiring more effort while being less profitable. I thought Sean pulled the wrong rein when they dropped chip supply because it excluded a huge number of potential customers, but it seems that the penny has now dropped.

If pre-orders are to be in 2023, we should see some income early in 2024. Let's hope we get a premium on the chips. After all, we will be substituting for Nvidia.

I wonder who will be supplying the model libraries?

The libraries will be determined by the specific applications, and will need to be adapted to Akida's format.

Libraries can be built up by training Akida, or, if existing libraries are suitable, they can be converted to Akida format (and supplemented by on-chip learning if needed).

Selling physical chips is less profitable???
Sure the margin is lower but say 30% profit on a $20 chip seems far more profitable to me than 100% margin on a say 40cent IP royaltie?????

 

[Pom down under]

Selling physical chips is less profitable???
Sure the margin is lower but say 30% profit on a $20 chip seems far more profitable to me than 100% margin on a say 40cent IP royaltie?????

Desperate times seeks desperate measures

 

[Diogenese]

Selling physical chips is less profitable???
Sure the margin is lower but say 30% profit on a $20 chip seems far more profitable to me than 100% margin on a say 40cent IP royaltie?????

Hi Justchillin,

It's a matter of scale. Before the switch to IP only, the company had decided that for orders in excess of 1 million chips, they would offer the IP licence, so presumably they had set that as the cross-over point, and maybe, for a startup, in switching to IP only, our eyes were bigger than our ... or is that pelicans?

 

[Damo4]

Selling physical chips is less profitable???
Sure the margin is lower but say 30% profit on a $20 chip seems far more profitable to me than 100% margin on a say 40cent IP royaltie?????

The key is the scale. VVDN's boxes will not be the same as Renesas' products for instance.
Sean also touched on "seven figure" License ($1m+) and at least 3x-5x in royalties ($3m+)

But as far as your calculations, it's volume that matters.
On a not-to-scale basis, if Renesas sold 100x products, more than VVDN boxes. you can see how things add up quickly.
Also I think the claim was nearly 100% so I did 95% for to cover any averaged ongoing costs related to supporting the client.

30% x $20 x 1000 = $6000
95% x $0.40 x 100,000 = $38,000

 

[Pom down under]

So am I reading this all right, we have now gone back to supplying chips? Or have we always been supplying chip? Or did we try to run before we could crawl?

 

[Justchilln]

The key is the scale. VVDN's boxes will not be the same as Renesas' products for instance.
Sean also touched on "seven figure" License ($1m+) and at least 3x-5x in royalties ($3m+)

But as far as your calculations, it's volume that matters.
On a not-to-scale basis, if Renesas sold 100x products, more than VVDN boxes. you can see how things add up quickly.
Also I think the claim was nearly 100% so I did 95% for to cover any averaged ongoing costs related to supporting the client.

30% x $20 x 1000 = $6000
95% x $0.40 x 100,000 = $38,000

3x-5x royalties……. I remember him saying that but didn’t understand what he meant…… I’m
Guessing he meant annually?

 

[Damo4]

3x-5x royalties……. I remember him saying that but didn’t understand what he meant…… I’m
Guessing he meant annually?

I think it was in Noah's podcast/interview.
And I don't recall a timeline so I would have to assume in total.

EG.
- $4m in upfront costs from Megachips (I forget how many nodes)
- $12m - $20m in royalties expected for that license.

The goal is obviously multiple licenses or, node upgrades, or even Generation upgrades that I assume cost more.

 

[DingoBorat]

Selling physical chips is less profitable???
Sure the margin is lower but say 30% profit on a $20 chip seems far more profitable to me than 100% margin on a say 40cent IP royaltie?????

Selling physical chips requires huge capital outlay, in the 10s of millions of dollars (at scale).

Capital we simply do not have, without huge additional dilution of the share float.

Which is why I've always liked the Company's decision, to go the IP sale route.

It hasn't gone nearly as nicely, as planned, but we do have Renesas and MegaChips (and associated sub licencees).

I think the Company's main focus, will remain on IP sales and I think we'll get more soon.
Additional smaller revenue streams, like VVDN, are more designed to kick start this Behemoth into motion, than a true change in direction, in my opinion..