Rise of AI Conference 2019 · Berlin

How to Turn Your Ideas in AI Into Products

Ulrich Schmidt · Segment Manager, EBV Elektronik

At Rise of AI 2019 in Berlin, Ulrich Schmidt, Segment Manager at EBV Elektronik, a semiconductor distributor, explains how his company helps customers turn AI ideas into products running on real hardware, from microcontrollers up to FPGA-based accelerators. He describes EBV as a roughly two-billion-dollar distributor of active semiconductor components with about 1,000 employees and 370 million parts in inventory, serving mainly mid-sized industrial, automotive and smart-consumer customers across 59 offices in Europe, Africa and Russia. He surveys where customers apply AI, including retail shelf monitoring, manufacturing predictive maintenance, medical image diagnosis, financial anomaly detection and assisted driving, then walks through AI-relevant product lines from three key suppliers: NXP's edge microcontrollers and crossover processors, Xilinx's FPGA accelerator cards for data centers, and STMicroelectronics' toolchain for converting trained models onto microcontrollers. He frames EBV's own role as focused on edge inference of pre-trained models rather than the cloud-based training side of AI, closing with a short Q&A on why dedicated AI chips remain limited among the distributor's supplier lineup.

Key takeaways

  • Schmidt positions EBV Elektronik as focused on helping customers deploy pre-trained AI models at the edge, not on the cloud-based training side of AI.
  • He gives predictive maintenance, such as detecting a failing ball bearing from sensor data before an unplanned shutdown, as a concrete manufacturing use case.
  • NXP's low-cost "crossover" processors, priced around three to four dollars, are highlighted as well suited to running edge AI algorithms.
  • Xilinx FPGA cards such as the U200, U250 and U280 target data-center AI acceleration and are already used in AWS's F1 cloud instances.
  • STMicroelectronics' X-Cube-AI tool converts trained models from frameworks like TensorFlow, Caffe or Lasagne to run directly on its microcontrollers.

Detailed notes

EBV Elektronik and its markets

Schmidt introduces EBV Elektronik as a distributor of active semiconductor components, part of the Avnet group, which also includes Silica and divisions selling passive and electromechanical components and server-rack parts. He describes EBV as handling the full customer journey from design and engineering recommendations to sourcing and logistics, comparing its logistics role to Amazon for active components. He gives EBV's scale as roughly a two-billion-dollar company with about 1,000 employees, just over 100 technical engineers, about 200 field sales staff, and an inventory of about 370 million parts held in a warehouse near Munich, with headquarters also in the Munich area. Its main suppliers include NXP, ST Micro, Microchip and Renesas, and it sells across 59 offices in Europe, Africa and Russia, primarily to mid-sized industrial customers, followed by automotive, smart consumer, smart grid, and high-reliability healthcare and medical segments; larger customers, he notes, are typically supplied directly by the chip makers.

Where customers apply AI

Schmidt surveys application areas he sees customers exploring: in retail, monitoring whether shelves are empty and using sentiment or demographic analysis, such as gender and age, to customize digital displays and advertising; in manufacturing, predictive maintenance that uses sensor data to detect, for example, a ball bearing likely to fail within weeks so maintenance can be scheduled before an unplanned shutdown; in medical imaging, helping flag cancer risk from diagnostic images to advise a doctor; in finance, detecting anomalies in fast trading; and in automotive, assisted driving, where he describes modern cars as effectively data centers generating large data streams that need processing.

NXP: edge microcontrollers and crossover processors

Schmidt describes NXP's range from Cortex-M4-based microcontrollers up to multi-core devices with as many as 16 A53 cores aimed at gateway and network applications, plus GPU-equipped devices for graphics tasks such as decoding video up to 4K. He highlights NXP's "crossover" devices, a hybrid between microcontroller and microprocessor priced around three to four dollars and running from 600 megahertz up to about one gigahertz, which he says are well suited to running AI algorithms; NXP is also adding dedicated IP blocks to accelerate AI functions and provides a software ecosystem for deploying pre-trained models. He notes EBV's presented solutions focus on edge processing rather than training, converting and running pre-trained models on-device for post-processing of sensor data.

Xilinx: FPGA acceleration for the data center

Schmidt presents Xilinx, comparable in size to Intel's programmable solutions business, as targeting data-center applications such as fintech fast trading, database acceleration, video transcoding and machine learning, with FPGAs allowing algorithms to be tailored and quickly adapted as new, more efficient neural network designs emerge. He describes the U200 and U250 PCI Express accelerator cards, the U280 with added high-bandwidth memory for memory-intensive workloads, upcoming network-focused cards, and a forthcoming 7-nanometer chip called Versal, combining universal and versatile building blocks optimized for AI. Xilinx's ecosystem also offers model pruning and guidance on partitioning workloads between standard x86 processing and the FPGA; he notes AWS's F1 cloud instances already use Xilinx FPGA cards to accelerate AI workloads.

STMicroelectronics: converting models for microcontrollers

Schmidt describes STMicroelectronics' Cube ecosystem and its X-Cube-AI extension, which takes trained models from frameworks such as TensorFlow, Caffe or Lasagne, converts them, and integrates them into ST's development environment to run on its microcontrollers, spanning entry-level Cortex-M0 devices up to Cortex-M7 devices running roughly 300 to 400 megahertz with extensive peripheral support.

Q&A highlights

Asked why dedicated, application-specific neural chips seem hard to find outside obscure Chinese suppliers, Schmidt says the answer partly reflects EBV's own supplier lineup, and that he is not fully familiar with competitors' offerings; he notes Nvidia is typically used for training workloads that need floating-point operations, while inference workloads can often use smaller bit sizes, and says that for most of EBV's suppliers, AI capability is currently built into their existing product lines rather than offered as dedicated standalone chips.

Questions answered in this talk

What kind of AI hardware does EBV Elektronik focus on?

According to Ulrich Schmidt at Rise of AI 2019, EBV Elektronik focuses on edge processing of pre-trained AI models, from low-cost microcontrollers up to FPGA-based accelerator cards, rather than the cloud-based training side of AI.

What is an example of predictive maintenance using AI mentioned at Rise of AI 2019?

Ulrich Schmidt described using sensor data to detect, for example, a ball bearing that is likely to fail within weeks, allowing maintenance to be scheduled ahead of time rather than reacting to an unplanned production stoppage.

How does STMicroelectronics help developers deploy trained AI models on microcontrollers?

At Rise of AI 2019, Ulrich Schmidt explained that ST's X-Cube-AI tool, part of its Cube ecosystem, converts trained models from frameworks such as TensorFlow, Caffe or Lasagne and integrates them into ST's development environment to run on its Cortex-M microcontrollers.

Mentioned

edge AI inference, predictive maintenance, FPGA acceleration, X-Cube-AI, crossover processors

Keep exploring

More from Rise of AI Conference 2019

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 1,980 words

Show the full transcript

the warm welcome from my sight before introduce myself who knows about EBV who ever heard the name EBV before Oh at least one basically what what we are we are distributor for active semiconductor components and basically we're part of the F net company which is divided into different speedboats where along with silica one of those who sell the active semiconductor components and then we have other divisions that sell passive and electromechanical components as well as parts that go into inter server racks for example and what we are doing is we're basically taking or going to the customers and from their ideas we basically follow the whole process whether it's the design the engineering we do the recommendation for for the right parts to choose the sourcing a big part of it is the logistics part of that I like basically Amazon we are doing for the active components and as well for the production at the end we're about a two billion dollar company about 1,000 people 100 a bit more than 100 in the

technical field so engineers that that can advise about the the products itself about 200 people in the field sales people and the others are backend or taken care of the whole logistics we have an inventory of about 370 million parts in our warehouse which is still in the Munich area and the headquarters also based in pouring a little bit west of Munich this is list of the of the suppliers we have in our line card predominantly it's nxp filings st micros micro sim a microchip as well as Renaissance that are currently quite active in this field of AI and I'll get to that later on in more detail and it's not just basically the CPUs or processing units it's also the sensors like image sensors pressure temperature acceleration that comes from from different suppliers which can be used in your products for yeah making use of AI algorithm at a later stage our segments the majority in terms of the markets is in the industrial area that means with so we we sell typically to

the mid-sized companies but the big ones that here ones are get directly supplied by the suppliers and we basically provide the services to the mid-sized companies throughout throughout Europe the next thing after industrial is automotive followed by a smart computer smart consumer and Smart Grid applications and then high rel and healthcare medical and for the for the main product groups about 30% of our of the products we sell is in the analog field so everything which is operational amplifiers power supply stuff then it comes processors which could be MCU up to a multi-core MPU optoelectronics power discretes memories of course that you need for for these MP use to do things and FPGAs which are or play a quite big role in this whole ai stuff as you will see we have a presence and 59 offices throughout Europe as well as Africa and and Russia this is quite a busy slide but it should give you an idea of where AI can be applied and where basically we see that customers are are thinking

about using AI algorithms in their product whether it's retail to for example see are the shelves empty what is the behavior doing kind of an analysis to to have displace or customize this place using sentimental analysis whether it's female male aged about so you can basically customize your your advertisement it's in the and the manufacturing a big field is this predictive maintenance where you can basically figure out by doing um I didn't find these the sensor data whether it's a ball bearing which is about to to break in with within the next weeks a month so you can do a timely maintenance on these on these machines when for example they have already a downed a plant downtown rather than when there is some some accidental problem happening so this is one of these examples where you can use I I for example in medical all these diagnosis of the of the of the big picture data to for example identify whether there is a risk of cancer she advice basically a doctor these are

things financed for for these fast Trading's to to detect anomalies as well in this field so you see there is a huge range automotive of course for this assisted driving this a big play I think the the cars in the future are like like data centers at the moment you'll find huge data streams going back and forth within a car so there's lots of lots of data to process and for that purpose basically nxp one of our supplier offers on the on the lower end MC use based on cortex m4 and then ranging up to multi-core a53 up to 16 a 53 cores in such a device mainly targeted for for gateway application or network application and then here in this area this is basically more graphic centric they have GPUs inside and also the proper interfaces to for example decode video data up to 4k and also do or to display and then there's an ya kind of what what they call crossover it's a kind of mix between MCU and amp use because

it's a fleshless device but it's typically in the in the price range of three to four dollars and ranging up from 600 megahertz up to one megahertz in processing capabilities so it's it's already four for this class of devices really great devices to use and to run your your algorithms on in particular here the next generation at the moment these devices are just multi core ARM cores basically but NXP is in the process to add dedicated IP blocks within these chips to accelerate in particular ai functions so yeah interesting stuff that is coming and as well they provide a a kind of software ecosystem to help you applying the AI algorithms and all these solutions I'm presenting at the moment are basically targeting the the edge processing side of things so we don't necessarily focus on the on the learning aspect of AI because this is typically happening in the cloud we rather with these devices we focus on the other pre-trained models that get converted and built into the software ecosystems of the of

the suppliers to finally run it on on the on the devices and use the the data that basically these products get that it will fit in with to do the proper post-processing second big supplier we have in our portfolio is silence silence is a think along with Intel PSG solutions programmable solutions the second biggest or yeah comparable in size for the FPGA business and basically they identified these different applications for this data center whether it's FinTech to to process fast ratings database applications to speed up databases for example to video processing whether it's a transcoding encoding decoding high-end processing just the pure machine learning because that's the beauty of the of these devices basically you can really tailor the your algorithms and run it on these FPGAs whether it's a new algorithms whether it's a new algorithm for example when new new networks are are discovered that basically use a smaller bit size this is something you can really adopt pretty quickly on on these cards this is the product portfolio and we have it

on display on our booth it's you 200 and you 250 these are just PCI standard PCI Express acceleration cards basically you can use within a PC environment or in a server tab of an environment you 280 comes with a additional HBM memory high bandwidth memory so for memory intense applications this is this is the right product to choose as well as upcoming products which which are basically for for network applications to do the packet inspection or whatever you can think of to speed up your tasks and then by the middle of this year there will be a new new generation at the top node silicon node and manufactured in seven nanometers called verso which is a combination of Universal and versatile that got again building blocks inside which are specifically tailored to optimize AI algorithms also from from starlings there's a huge ecosystem targeting the software aspect they they can offer different different models do some pruning which means do some optimization of neurons that basically don't have a real impact to the to the

neural network and then on the left side we have it basically again for the embedded application which means devices which have multiple arm course along with the programmable solution part of it the FPGA and on the right hand side software called software-defined acceleration basically where you can run your algorithms shrew and basically the two gives you some guidelines whether the algorithm or parts of the algorithms are better partitioned to run on the on the x86 for example and while others are basically more shifted towards the program else programmable solution which means the FPGA are part of it there are different development boards that are available you can basically start playing with and there's already different cloud providers that utilize these cards for example IWF AWS f1 is using FPGA cards for accelerating AI algorithms you can run in the cloud the third in the list is more towards the lower end type of of applications in the AI space they have a eco system which is called cube to basically configure the device set up

the proper efforts and there is an extension to it called X cube AI basically to integrate into this integrated development environment and it's again like with nxp takes different trained models whether it's tensorflow cafe or lasagna do some conversion and integrate it into into the tool to later on run it on on the MCU and SG is one of those suppliers which have a really big portfolio ranging from entry class cortex-m 0 devices up to the cortex and 7 devices that typically run in the range of 3 to 400 megahertz come with a huge set of peripherals to to help you interface into the to the world so there are lots of devices you can choose from and basically EBV is is here to help you guys selecting the proper device for that come to see us AI gets real so we get a lot of feedback during this session basically to see really where AI is really processed because typically AI is a just mathematical operation taking a value and and multiplied with a

certain factor you do that in parallel many many times and you can really see where AI gets processed so come to see us if you want and if there's any questions it's now time for that yep I've just checked to find some neural chips on the net but I only found some obscure Chinese suppliers to provide new chips why can't we see any application specific it depends first of all we have this selection of suppliers on our line card so I don't really follow completely what what's what's available within our competition of course Nvidia is one of the big ones that typically get used for for all these learning processes because there you need a floating-point operation whether or raw the the interference type of thing typically can can get along with smaller bit sciences so yeah these supplies typically focus how a I fits into there they're kind of base product lines so this is not necessarily a dedicated product offering they have envision at the moment at least from from what I know

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.