Nobody forges their own screwdriver: Inside the chip-down versus module debate

Ralf: Actually, I’m business development manager for embedded modules, custom boards and system level products. And I’m driving the technology parts, being the first interface to the technical departments of the customer, trying to understand their requirements and then coordinate everything internally with the back office, with the R&D department. We are having our own development with the footprint in Germany, as well as our own productions in Germany.

So we really know how to handle this business. And the interesting thing is getting this feedback from the market, summarise this, focus it on this, seeing the trends on the market as well. And then, because of the fact that we have no longer our own sales force, contacting our partners — in this case, it’s Silica.

It’s a quite interesting market segment. The agriculture and commercial vehicles at the moment are really challenging. Having AI on board, having connectivity. We see in our discussions with the OEM that they have to face three things. First of all, supporting the existing machineries in the field. Then identify and understand today’s requirements based on AI, based on data volumes. It is no longer just data transport, it is just fleet management. It’s just securities on the over-the-air updates, fuel efficiency monitoring, all the things. The challenge is the complexity of the system level itself first, and having also the view on the future architectures, which is coming soon.

And then, of course, having also the discussions: how can we handle this? And for us it’s clear, it’s modular concept. So modular concepts are no longer a technical discussion. It’s just really a business decision to do this.

Frank: That’s true.

Ruth: And Frank, what about you?

Frank: Yes, I’m still more than 25 years within the company. Started as a sales engineer and changed my role from a customer-related to a product-related one. I am responsible for a product group, which is the complete embedded solutions — from not only Tria, also displays, memories and stuff like this.

Ruth: So let’s define terms first. What does CAV actually mean in your world, and why isn’t it connected and autonomous vehicles?

Ralf: Well, actually CAV, in the traditional way, was just construction machineries and agriculture vehicles.

Ruth: Okay. There’s so many abbreviations in the market and in every industry, and everybody defines their own.

Ralf: But, as you can see, we call it commercial and agriculture vehicles, because of the fact that we meet very often the same requirements in sub-applications, like public vehicles, like trains, like buses.

Wherever you’re just discussing, you’re having every time the same story. Having the connectivity as a must-have discussion, as well as, of course, the massive beast of AI power which you have to handle. And therefore they just meet and smile into one request, more or less, that we say: okay, commercial vehicles also is a part of this construction.

What was in the past only construction machineries was harsh and robust, secured applications. This just coming together. And we, as thinking of platforms, we are willing to bring into the market proof of concepts which fit whether in the first or the second application styles — doesn’t matter.

And we try to cover everything, having a deep dive with the customers. Look here, we nearly cover, with our understanding of the market, 80% to 90% of your requirements. Let’s discuss now the differentiator, the last 10%, to bring you right on track.

There’s a clear trend to see this, that this is a massive impact in this. I think, Frank, you agree on this as well, right?

Frank: Yeah, that’s right. It’s a huge market. Usually, if you take only construction and agriculture, this is only a small market, and if you add commercial and agriculture, the market is a little bit bigger than.

Ruth: And you already mentioned that AI is a big factor in what’s driving the CAV market right now.

Frank: Yes. And we will see also new concepts in the future. So imagine, in the moment you have ChatGPT or Claude as a cloud solution. And in the future we will have this as a chip onboard. So you can talk to your machine like you talk to other people. So we will see a change in giving access to machines.

Ruth: And functional safety keeps coming up in this space. Why does it matter more here than in consumer electronics, for example?

Ralf: Because of the fact that you are just acting in the field where, for instance, if you drive a tractor through a field and there’s a human being in front of it, it has to automatically stop. Otherwise you hit an accident, which nobody really wants to have.

So therefore, this is a big, big issue to handle this AI. Of course, it’s this right supporting tool. It’s a massive piece to handle this, to programme this in the right way. So software engineers having really a challenge. But this requires as well — and this is much more difficult than in the consumer part — real-time applications.

So this means you have sensors, you have actors working around this in your infrastructure, around the tractor, for instance. You have to control everything, that the entire data has to be summarised, interpreted, and then having the actions doing in the field, by being a co-pilot for the driver itself.

Frank: Yeah. Hydraulic systems must be controlled to not harm any other people.

Ruth: I suppose there are also certification requirements.

Ralf: Yeah, of course, there’s special certification required, case by case. It’s a bit different, but more or less it’s just meeting all the same. So having real-time applications, having also secure, protected fallback solutions, let’s say, in case that the human being makes a missed decision or whatever — that that is controlled on this. That’s correct.

I think, Frank, you have a good software tool for this, right?

Frank: Yes. We will have solutions for this, and we will see this within a timeframe of approximately 12 months latest, to see products in the market will help this getting solved.

Ruth: What AI-driven autonomy features are we talking about here?

Ralf: Just the positioning of the vehicle itself. Just also controlling, for instance, some drones just around you, to water the field, for instance, to analyse the ground — all the things you have to control.

And then, of course, this is really, really good to understand this, having an efficiency. So the efficiency means: me as a farmer, I want to have the best economical balance in my system. So how much fuel do I need to drive the tractor? What is the watering system doing in the field? What is the harvest itself? Which chance do I have to get a good one?

In parallel to this, the driver wants to have, of course, the much more comfortable environment on place, and having these discussions as that. This is not only data recording and moving this from A to B, it’s just an interactive session which you have to control. And in parallel on this, if just also control the actors and sensors around this, you have to take care on the predictive maintenance.

Means, if a sensor may fail in the field, you have to control this, because if a machine is standing on the field, it costs money. The originally thinking — yeah, we can double-check this in the evening, whatever. No, it’s gone. It’s really gone. It’s just online doing this. Overall machine health management tools control everything in millisecond. They got the information: hey, this sensor may fail, replace it, whatever, to be more efficient on this.

Frank: On the other hand, we also have solutions who are controlling the pesticides amount. So reduce it and only spray it if it’s needed to.

Ruth: Yeah. So this is what I find really interesting about commercial and agricultural use cases: that autonomy isn’t only about driving, it’s about so much more. It’s about safety, it’s about crop management, harvesting efficiency. That is what I find really interesting.

We already bit touched on some of the use cases. Let’s get concrete. Walk me through a special use case you would like to share. What are you seeing in the field today?

Frank: Yes, as I just explained, so we have a pesticide control, which is very effective. And this is a camera-based system, which is mounted on the tractor, and it looks if it needs to spray a pesticide or not. So, and this will do it very, very good.

Ralf: For instance, you have also the discussions, if you’re just looking into the US for instance, they have massive hectares of fields, which nearly 100% automatically driving themself, the drivers. So they’re just giving the request: hey, I just need more fuel. Then they’re coming and bringing it on the field.

So they are just more or less in case that there is something onboard, but originally they can drive autonomously. Having these actors, as we all know, with just getting the first discussions also in our cars, having driving assistance, whatever. This is exactly the thing which is working there, much more challenging cases because of the harsh environment. But it’s just a step before the standard consumer, because in the field there are not so many human beings having an accident, whatever. But they are more or less the pioneers on having this also working in the standard automotive sector.

Ruth: I think in our briefing call you mentioned a use case. It was about autonomous milking systems, I think. And didn’t you quote the farmer saying he’s not owning a farming company, he’s now a robotics company?

Ralf: Yeah, that’s correct. So there’s just — the original farming was just also control everything, feeding the cows, whatever. And now this just more or less, they’re having just an app on their mobile and seeing exactly, they’re having cleaning robots, feeding robots, to get the cows just satisfying in their infrastructure.

They’re just mowing the grass itself in the special way they want to have this. They just produce everything, package it by packages for the cows, and doing everything nearly automatically, and they just only control this. Yeah. Even if something went wrong, you need a doctor or emergency cases, they just get the alarm and get the call: hey, there is something, you have to come to the farm and double-check this. This is just a bit more relaxed.

When I was a child, a farmer wakes up three to four a.m., having short breakfast and going out. Today it’s more relaxed. Some farmers are more or less having this as a part-time job, which is interesting. Not just saying they are just lazy and sitting there, blah blah, and don’t know what they have to do, but it’s just really supportive on this.

So, but this is also, again: being a farmer today, you need to have a real IT infrastructure. Means you need servers to store your data, to make an interpretation. What is the efficiency on this? What can I do? It’s just a business case, like a company, to make more efficiency on this - instead of just looking out, oh, what’s the weather forecast? I’m not sure. Ah, let’s see, my experience tells me this or this or this.

Ruth: We will take a short break. Stay with us and we will be hearing from our guests very shortly.

This podcast is brought to you by Avnet Silica, the Engineers of Evolution. Subscribe to our Avnet Silica newsletter or connect with us on LinkedIn. If you want to learn more about us, we have put information and links in this episode’s show notes.

Ruth: What other industries are profiting from this? Is there a use case maybe outside of agriculture you could share?

Ralf: It’s massive. For instance, as I said, the public vehicles, they’re using this as well. Having cleaning machineries for streets, whatever, or even the standard train and bus application itself. Formerly called — we just had this as transportation, more or less, because human being transportation. But this is also something they’re having, for instance, the new challenge: having people counting systems. Means there’s a camera system of the vehicle itself, the bus or whatever, and counting the people coming in and getting out, that there is not too many passengers onboard, how to handle this.

And also in case of having a criminal case, where the government wants to know exactly what happened there, where it’s just to follow the route of this criminal guy. He’s using the bus line A, B, C, whatever, and to the next stop going out, there changing, by having a mask, his outcome, whatever, is different than he is getting onboard the bus. They have to analyse this, they have to figure it out, and to serve this to the police, whatever, to get a better understanding on this.

So there’s also much more intelligence coming in these vehicles. There’s not only construction machineries, mining.

Ruth: Mining is an interesting use case, I think.

Ralf: Yeah, exactly. This is the next thing. It’s just getting more and more impact. Everyone has more or less the same requirements.

Ruth: What’s the use case there?

Ralf: Yeah, having a coal mine, for instance. In the past, the guys were under the ground and digging deep to get the coal out of it. Today it’s just robotics, bringing out the coal and transport it through the survey. This is not the difficult as it was before, having this industrial revolution, let’s say.

Ruth: That sounds very promising.

Ralf: Once again, you have to integrate this massive beast of AI technology in robust and secure design.

Ruth: What do developers in the space tell you keeps them up at night, when trying to tame the AI beast?

Ralf: This is exactly the thing. They’re also having some fallback solutions internally on this, controlling everything. But I’m now 19 years in this field. If I just compare the requests coming for a project 10, 15 years ago and today — it was, in the beginning, hardware driven. Hardware driven, and software: yeah, we will handle this with an operating system, all good.

Today it’s more or less not only 50/50, the impact is nearly 80% in some applications based on software. And today the decision is made: we have the software request, A, B, C, we have to handle this. And based on this, we will calculate what kind of hardware support do we need to have this AI running.

So in the past you said: what do we need? Okay, we want to have an efficiency system level, power consumption, all the things which is best, wonderful, passive cooling, whatever, in the embedded application. All good. And then we just implementing the software.

Today, the software is the first step we have to discuss, because the requirements will covered by the software. And then they make the decision what is the best fitting solution on hardware side: having a platform design and carrier board, where all they have the interfaces on it, and having the scalability through standard embedded module — different core sizes, memory sizes, all the things just having impact.

You have the scalability on this, you have also the support for the former solutions in the field, but also having the perspective to the future-driven things, like changing the module itself, having the next generation of CPU on it, memory on it, and everything’s working more or less fine. Of course, the software you have to handle, but on hardware side, you have a stable system for at least 20 to 25 years.

This is the thing. And this gives us a big impact as a manufacturer of embedded modules, as a design-in partner, where we are seeing us with these OEMs, to give them the support to handle all these three cases in one.

Ruth: So I’m moving now to the design question. What does chip-down actually mean, and why would an OEM choose it?

Frank: Imagine you’re working in a big industry and you need a screwdriver, and then you start to forge it. Yes, and build it by your own. This is exactly what is happening in the industry. Everyone is doing their own system, chip-down system by its own. So, and they will not use a ready tool.

So what we offer, what we’re able to offer for our customers — and this is what’s changing in the moment, due to memory needs. And this will not be possible for everyone to design their own system in the future. So this will be so complex and expensive, even for the tools you need to develop such a system. So therefore, it is in the future that we will see more and more module solutions.

Ralf: That’s correct. And the other thing is, if you’re just making a chip-down, you define the requirements and you make the design, you solder it down in the SBC, and that’s it. The next generation, you have to make a completely new SBC. So completely new PCB itself. This is exactly the thing: reduce this effort, the design risks, the costs, short time to market, all these things.

Our slogan now, as said, and modular concept is no longer just a technology choice. It’s more or less a business strategy. Having one platform design — of course, it costs a bit more originally for an SBC compared to a two-board solution, there’s the gap of 10%, something like this. But the automatization is based on the platform design, on the value projects you will support on this.

So means, having an existing design supporting this, the next future platform is automatically designed in, because you have to switch the heart of the system, the module, and then, seeing, by adapting the software, the next result. Immediately, the look and feel is the same. The end user has trust in the platform, which they have for 10, 15 years, whatever.

Seeing now the efficiency: oh, this is a brand new one, powerful, great, more intelligence in this - but it’s still the same infrastructure, what they just visibly had. So the housing is the same, the style, the positioning of the machine, everything is the same. There’s nothing different. And replacing this, it’s quite easy doing this, instead of: okay, we have to reconstruct this, design a completely new one, try to fit it in, and all these things.

So it’s also from maintenance point of view much more better to sell to the end customers. Because they need trust in their existing platforms, or they have this, and they need this for the future platform as well. So the technology migration is much easier to sell in platform thinking with modular concepts. Less risk for the design and the cost. And it has, of course, based on this, massive better time to market.

Ruth: Okay. And from a distribution side, where does the scalability actually pay off for an OEM? Is it cost, is it lifecycle, is it something else? Is it all of the above?

Frank: I think it’s a bunch of things. Yeah, of course, time to market is one important thing. On the other hand is, as I mentioned, you do not need to develop every tool by your own. So in the future, as mentioned, this will be more and more a module solution than chip-down.

Ruth: So does it generally depend on the vehicle and the volume?

Ralf: There’s an impact, of course. You won’t have a completely customised system level product with having only a few hundred per year or whatever, because the development for this and the cost are too high for this. But honestly, with the OEMs where we talking about this, there’s no fear about this. So we are not afraid to having the right business case on this.

The challenge is, again, seeing the more intelligence in the system, which is needed, of course, and making this scalable, sustainable and available for the next generation of the machine. This is the challenge, what we try to cover, and where we see the best solution is really having a modular concept.

Of course, we can also do, because we have our own development, just using the IP on the platform design for making it as SBC. But the trend itself is really having modular concepts, to changing the module itself.

We have a special trend at the moment. ARM is getting much more performant than x86 in the compare in the past. So the gap is getting more and more thinner in between. The performance of ARM is much higher.

There’s a new future trend. We are having some SMARC standards, which is a stackable module itself with an edge connector, which is fine. The new trend is now - and this is also an impact on having the discussion whether you make a chip-down design or module concept — is an OSM module itself. It’s a soldered-down module, where you have also the scalability on this, with the character in the system of having an SBC. So it’s an LGA design, which you just solder down on the carrier board itself, which you can also, having the next design, just changing the module, the OSM module itself, having the same design, but you don’t have to do the completely SBC brand new.

So the design you can cover, everything is fine. Replacing the module, the core IP itself, replacing this, having the software impact, and then you still have a character of an SBC, as formerly only happened within chip-down, and having also the modularity on this. This is the new trend, what we see, and the performance which is coming there.

Some manufacturers having also AI generated the things on this. Quite interesting.

Ruth: Before we close, is there anything I haven’t asked you that you wish I had asked you?

Ralf: What is very important to understand is: machinery from today is no longer just powerful, there is an intelligence in this. We are facing three main challenges out of this. AI and connectivity creates for us the opportunity to discuss with the OEMs. The system complexity creates really a challenge for both of us.

But finally, we have the solution in our point of view, and we practise having long customer intimacy with customers for more than 20 years. Where we originally discussing making SBC based on our IP, they won’t have this discussion anymore. They just want to have a modular concept. This is the point.

So modularity is, as a system strategy as well, really a business strategy, no longer a technology choice. And therefore we just having the solution out of this, and more and more discussions with the customers. So let’s say, in 99%, nobody’s really asking for an SBC anymore.

Ruth: If you had to put together a soundtrack for this episode, what song would you put on it?

Frank: Hermes House Band, “Country Roads”.

Ralf: I’m a bit more technical driven, to be honest. For me, it’s Daft Punk, you know, “Harder, Better, Faster, Stronger”.

Ruth: Oh, that is—

Ralf: This is really good. And actually there’s a good version, the radio version from 2007.

Ruth: Okay, I will keep that in mind.

Ralf: They just mix “Around the World” and then “Harder, Better, Faster, Stronger”.

Ruth: That’s terrific. You know what is really funny? I had a guest from AGCO Fendt, and guess what song he picked?

Ralf: The same.

Ruth: Yeah. Isn’t that interesting?

Ralf: Yeah, makes sense.

Ruth: So the industry is in unison about the soundtrack for agricultural vehicles.

Frank: Thanks for having me.

Ralf: Thanks all for having me.

Ruth: Thank you for listening to We Talk IoT. Stay curious and keep innovating.

This was Avnet Silica’s We Talk IoT. If you enjoyed this episode, please subscribe and leave a rating. Talk to you soon.