#452 – An Interview with Kieran O'Leary

Kieran Oleary: This is The Amp Hour Podcast. Released July 28th, 2019. Episode 452. An interview with Kieran O'Leary.

Chris Gammell: Welcome to the App Hour. I'm Chris Gammell of Contextual Electronics. Hi, I'm Kieran O'Leary and my company is Mixed Signal Systems. I'm based in Edinburgh, Scotland. Hey, Kieran, how are you doing? Yeah, great, thanks. Good to talk to you at almost 1am.

Kieran Oleary: Yeah, yeah, it's real late over there. But you were a night owl and that seems to work out well. You were filling in nicely for a cancellation we had. I really appreciate that. And I know you're from the consulting forum. So what kind of consulting do you do?

Chris Gammell: So the clue isn't the name, really. So the company name is Mixed Signal Systems. So our core competency is in designing high-performance Mixed Signal products. So products with both analog and digital elements. So that core strength allows us to design products for a wide variety of markets and applications. So, of course, at the moment, there's a lot of work in the IoT space, you know, with a variety of sensors and sensor fusion and so forth. But based on my background, we do a lot of work in the audio electronics space right through to kind of high-bandwidth multi-gigabit systems, such as data acquisition units and so forth. That sounds kind of tough. Well, in some ways it is. I mean, the way my career has gone, the bandwidth has gone down, but the bit count has gone up. If that makes sense. Yes. Yeah. You know, so, you know, like you might have, you know, a 12-bit video DAC or ADC, you know, a couple hundred megahertz bandwidth. But then you move into the audio space and the bandwidth comes way down, but the bit count goes way up. So, different challenges, different challenges for the different markets, but the physics don't change.

Kieran Oleary: Well, that's good. That's a good take on it. I mean, I like that, you know, you're keeping a clear head on that because it's like, that's, I mean, it seems, it seems like a kind of intimidating thing to have, to have these high bit rates or high bandwidth. What was the trade-off game? It was bit rate versus bandwidth or?

Chris Gammell: Yeah. Yeah. Like, so, you know, yeah. So, video DAC, 12-bit, you know, but a high bandwidth or higher bandwidth and then audio, you know, and the bandwidth comes way down, 20 kilohertz or whatever. But, yeah, the bit count goes way up. So, what is your background?

Kieran Oleary: Like, how did you get to this point in the first place?

Chris Gammell: Well, you know, just kind of back to the start. So, you know, I tell from my accent, I'm Irish, but so I'm from the south of Ireland, southwest of Ireland, County Kerry. And I went to university in a fairly small university called UCC University College Cork in Ireland. But it's got a good electronic engineering department there. But its claim to fame is George Boo, Boolean Logic, was the first professor of maths there or professor of mathematics. So, did my electrical and electronic engineering degree there. And maybe we'll talk a bit about it later. But my final year project there was a control engineering project. And I worked on the control algorithm for my robotic arm. And, you know, my classmate, colleague of mine, it did the power electronics. And everything worked remarkably well. Everything spread out in the bench with, you know, a nice little DSP running the control algorithm and all the power electronics controlling multiple motors for this robotic arm. And then the open day was approaching. So, we said, oh, we better make this robot look a bit nicer. So, we popped everything into an enclosure. And, of course, you know, it looked amazing. But on the open day, the robot took on a mind of its own and went berserk. And that's when I first learned about EMC and signal integrity. And I promised myself then, never would an EMC issue make me look like a fool again. But, of course, that's strictly speaking. It wasn't true. But I learned my lesson, you know, early, you know. And since then, I've always had an interest in that side of things. But I was a graduate in Xilinx in Dublin. Their European headquarters was in Dublin. And it was a really interesting time to be in Xilinx, actually, because it was the early 2000s, just after the dot-com bubble burst. And at the time, you know, it's still the case, but particularly at the time, an overwhelming majority of Xilinx and, indeed, Altera's revenue came from the telecommunications market. Now, there was some other areas like military and aerospace and all that. But for the most part, it was the telecommunications industry. And, of course, at the time, you know, it was a difficult time for, you know, for a lot of companies in that industry. So it was an interesting time to be at Xilinx and worked in the embedded applications team there as a customer applications engineer. So the early days, you know, I hear Dave, you know, talking about it often about, you know, with the FPGA is how there's more and more of an integration of hard IP, you know, and, you know, to go along with the FPGA fabric. And that was really around the time, the early 2000s, when they started that. So the PowerPC embedded in the Vertex 2 Pro. Oh, yeah. So that was their first device, you know, and the microblaze had just, you know, kind of come into being. And, you know, that was, God, it's a long time ago now, but it was kind of the natural evolution for, you know, for the FPGA fabric. Like, you know, you needed more than just, you know, a big, big, you know, array of lookup tables. You needed a bit more. And that evolution has continued to today. But, you know, there was a couple of years there in Xilinx in Dublin. And...

Kieran Oleary: Well, before, wait, before we leave the Xilinx side of things. Sure. So tell me a little bit more. So you were on the application side of things. I mean, so you saw this going into all these different places. What were... I mean, people were using it as a joint, as the embedded micro in there, like you said, with the hard IP? Yeah. So, yeah.

Chris Gammell: So I was in the embedded applications, you know. So at the time, I mean, it's very standard now, you know, it's completely normal nowadays. But at the time, it was quite novel that you would have maybe a telecoms base station or something like that. And the FPGA, you know, it could handle, you know, masses of data. Like the throughput would be very, very high. But you still needed a processor that, you know, didn't have to be particularly high bandwidth. But it needed to manage the routing of those transactions. So if you manage, you know, you have a voice over IP network or something like that. And you've got all these packets. And, you know, they need to go to various locations. And it made sense that the data pipe needed to be very high bandwidth. But the management of that data pipe didn't need to be so high bandwidth. But nonetheless, you know, it needed to take place in the system, somewhere in the system. And FPGAs, they're not cheap now. They certainly weren't cheap then. So perhaps it made sense to accommodate the cost of the processor, you know, within the FPGA fabric or, you know, within the system on chip. And there was lots of applications where that made sense. You know, robotics was an area where it made a lot of sense. And it was the type of application where those, you know, really kind of top-end FPGAs were able to, you know, to find a good place in the market where you had, on one hand, high bandwidth requirements. But on the other hand, you needed a management system to, you know, to make sure that that data was, you know, taken from the right location and piped through to the right location. And the evolution sense has kind of proved that strategy right. There's, you know, FPGAs, they're power hungry. You know, they're not, they don't have the absolute performance, perhaps, that an ASIC would have. But they've got other advantages that are fantastic in a lot of applications. And we see that now, you know, with all the various hard IP, with the multi-gigabit interfaces. Even quite recently, Xilinx brought out an RF SOC, which is quite novel, that had an RF, you know, or multiple RF DACs and ADCs integrated into the package onto the DAI. So that you can do, you know, entire software-defined radio or radar systems on chip. You know, you can do your DSP, you know, all this kind of thing. And that all started back, you know, nearly 20 years ago now, where, you know, like, the FPGA was, you know, a very high-performance CPLD. You know, we're, okay, we're GlueLogic and we're piping this from A to B. And, of course, you know, FPGAs were capable of much more. And then we just happened to be, you know, be there at the time.

Kieran Oleary: And it's crazy how that's moved in the RF space as well and how, I guess, the integration on a chip and, like, the changes in process have also allowed that as well. Because higher speed, higher performance type stuff, or even more exotic stuff that RF sometimes requires, being able to put that all on the same DAI is very, very impressive.

Chris Gammell: I think that's a fair point. I mean, there's a couple of things there to trade off. And the guys in Xilinx clearly think it's worth a trade off. You know, you've got, you know, an FPGA, you know, perhaps the noisiest thing electromagnetically that you might have on your board outside, possibly, you know, the power supply. So putting that, you know, right next to your RF front end is quite an interesting strategy. But there are good reasons to do it. I mean, there's nowadays something that people don't often think of is if you lay down, you know, an interface between two devices on a PCB, that can be quite power hungry, actually, when you're transmitting, you know, gigabits, you know, of data. And the bandwidth, you know, the bandwidth of the channel, of the interface between the two devices is quite high. You know, you know, there's capacitance on the track and everything, you know, as you switch your ones and zeros, you know, even a serial manner. That can consume quite a lot of power. And if you can localize that, maybe you can save some of that power. That can be advantageous. Of course, there's a space saving. I mean, what I think would be really cool is to take one of those RFSOCs, which have, you know, multiple RF DACs and RFADCs. And you've got, you know, a couple of processors in there and, you know, a big FPGA fabric. And you could do something really cool with a drone. You know, you could build your entire drone or UAV around this one device. You know, you could have, you know, your remote, you know, your remote control of the, you know, RC radio, if you want to, you know, think of it that way. Coming in over one antenna and you could have, you know, passive radar to see what else is around you. You could have your motor control. All these different things can be done in one chip. And of course, I mean, it might be the world's most expensive, you know, hobby drone, but it could be a really cool, you know, it could be a really cool. $15,000 drone. Yeah. And the battery, like, you know. You start weeping every time it crashes. Oh, well, yeah. Yeah. That's the thing. And, you know, typically you think of the drone motors consuming the power and you go, oh, well, this is kind of funny power profile here. You know, the processor controlling the motors is drawing as much power as the motors. But it's that level of integration where you could even imagine one system on chip doing the RF, doing the motor control, doing any kind of, you know, Kalman filtering or anything like that, you know, for any kind of three dimensional position estimation and stabilization. These kind of things all in one chip. It really is quite impressive. And you can, you know, the Altera guys might say different. But, you know, of course, but from a Xilinx perspective, you can trace that back to the Vertex 2 Pro, which was a Vertex 2 with, you know, a little bit of vision, perhaps. Mm-hmm. And that's great.

Kieran Oleary: Yeah. So that role was, it was an application engineer. And then you kind of kept doing that later on. I mean, we've had a couple of application engineers through here on the show. But, like, what is that experience like? And then how did that end up playing then into consulting as well?

Chris Gammell: Okay, yeah, I'm glad you asked me that because, you know, I've been a longtime listener of, you know, the Ampar and I really enjoy it. And one time I remember you were having some discussion about FAEs, field applications engineers who are, you know, typically really the, you know, the front line when dealing with, you know, with customers from a technical perspective. And Dave said something like, oh, you know, I can do without the FAEs, you know, like I know what I'm doing. And I was just, you know, screaming at the radio or whatever, like, oh, my God, I need to stand up for the applications engineers. This is a common thing, so I wouldn't. Yeah, yeah, I understand. But there, and it is, I mean, applications engineering is a broad church. And, yeah, you know, it's not typical, perhaps, like, that you would go from applications engineering into consultancy necessarily. So it is worth maybe kind of exploring that. And, you know, even talking among applications engineers, you know, like, how would you define your role? And it's one of those roles that can be difficult to define. But basically your job is, you know, to shepherd the design from the development organization, you know, in your company into the customer's hands and the finished product. And depending on the type of applications engineer that you are, it is your responsibility kind of as a team to ensure that from a technical perspective that the silicon or the device comes from, you know, your organization into the customer's organization. And that it works, that it works in the application to a standard that the customer is satisfied with. And that means a great, you know, there's a great deal of tasks that need to be done to ensure that that's the case. Like I said, the FEs, the field applications engineers, you know, they're out there day in, day out, engaging with customers, you know, working on technical solutions, you know, for the customer's application. And for smaller clients, you know, that aren't using, you know, the latest technology, it can seem like that the FE is just kind of a technical resource that it can do without, you know, if you know what you're doing, you know. But for, you know, strategic customers of a semiconductor company, the relationship with the FE is vital because, of course, they'll be dealing with engineering samples, brand new silicon. And there may not be a data sheet available. You know, the silicon you're working with may be engineering samples. There'll be a lot of errata. You might have a very strong relationship with a particular customer and you'll want to get feedback from that customer, you know, like they'll want their guide that they can pick up the phone and they can go, hey, look, we're just not getting to performance here. Like my background is in mixed signal products and maybe the digital side is fine, but maybe the performance of the ADC just isn't up to scratch. Now, part of that can be, you know, with a sophisticated ADC or DAC, you know, there can be, you know, quite a lot to configure them to get the performance that you expect. Or it might be that the silicon just isn't up to it and that that needs to be managed. And, of course, sometimes you think that a semiconductor sells silicon. Of course, it doesn't. It sells a solution to a customer's problem. And along with the silicon.

Kieran Oleary: That sounds like a consultant or someone who's selling their services pretty much.

Chris Gammell: Yeah, but it is very important because, you know, it is one of those things like, you know, you know, I have this for many years, you know, discussions with other groups within the organization. And, you know, they'd say like, oh, I don't really know what you applications engineers do. And they say, well, you'll know it when we're not here. And it really is a case of, OK, well, OK, the next time your ADC or DAC or whatever doesn't quite work when the customer rings you directly. And he interrupts your workflow and he says, well, we don't care, you know, like that, you know, you're working on the next revision of silicon. Like you need to get this silicon working and you need to help us understand how to integrate that into our platform and get the analog performance that we need. And, you know, it's there's a lot of a lot of effort goes into that, you know, writing data sheets, errata, all the way through to the more strategic stuff of like managing the relationship. And even, you know, next generation definition of next generation silicon or even to some degree involvement in pricing discussions. Of course, you're not a sales engineer, technical marketing guy, but you have to try to understand what the market will bear at a particular price point and so forth. And that does require technical knowledge.

Kieran Oleary: I think the thing that would be scary, scariest for me would be so many times I, you know, me as a consulting engineer or working as an electrical engineer at a company, there's always someone to ask a question to. There's always someone like I can be like, you know, assuming I get enough attention from a vendor, I can go and ask, you know, a local distributor rep or FAA distributor or moving on up to the actual FAA at that company or, you know, and then sometimes they say, hey, let me call the factory, the AEs, I can talk to them or whatever. And, you know, there's always someone up the chain I can talk to. But I imagine that in application engineering, like you can go talk to the silicon designer, but sometimes you're just like, I'm the end. I am the person that figures this out. So I'm sure that you've run into that a couple of times.

Chris Gammell: Yeah. I mean, of course, if the organization is working well, you know, you never feel on your own. But sometimes, I mean, it's just a business reality. And I guess perhaps perhaps that's the thing that maybe applications engineers sometimes, particularly more junior applications engineers relative to junior design engineers will have much more of an appreciation of the business case or the business environment. Now, as designers become more senior, they're exposed to more of that within the organization. But typically, an applications engineer will get that much earlier in their career at a more junior level. And you might be the end of the line in the sense that the business case just simply isn't here, you know, to, you know, to involve the designers, maybe to do a new silicon revision or to do the level of research required, you know, to, you know, to come up with, you know, a solution. It just may not be economically viable. And sometimes that can be a difficult discussion with a customer, but sometimes in conjunction with your sales guys, you know, that that is left to the applications engineer. That's what you're trying to do. You're trying to get the right product for the application. And actually, in many ways, that was a good grounding. You know, that was a good grounding moving later on into a consultancy role. You know, actually, there was some benefits to that. The understanding that communication is important. Trust with a customer. If you have a good ongoing relationship with your customer and, you know, one time it's not that you're letting them down. It's just unfortunately at this time, you know, at this time we can't deliver that particular requirement, you know. But please understand for the long term, we, you know, we intend to have a relationship with you and it will be in the next silicon or it will be in the next product or we will do our best for you. And all those experiences built up over the years as an applications engineer on site with customers or dealing with them remotely does give a perspective that perhaps would be helpful for somebody, you know, interesting in being a consultant themselves.

Kieran Oleary: I guess that I'm interested in that relationship you're talking about, too, because you're talking about like the key customers. So these are like large scale, big businesses that are going to buy millions of units. What was that actually like? Was that was that akin to like an FAE or was more like a handholding type of thing because it was so early, early days on early silicon or what?

Chris Gammell: So in my in, you know, before I went out my own, I was in three companies, you know, three semiconductor companies. So I was with Xilinx and later analog devices and Wolfson microelectronics. And in each of those companies, you know, we were dealing with strategic customers or tier one customers. You know, there's different terms for them, but people appreciate like, you know, where the majority of the money is coming from. And of course, they like any other business, like I mean, they get the premium support. So, I mean, those large customers, they will have FEs in any reasonably sized semiconductor company. The large, the very large customers will have dedicated FEs, you know, particular even within an organization, say like Apple or Samsung or somebody like Panasonic, the really big companies, particular subgroups within those organizations will have FEs assigned to them. And those guys are there on site every day behind the FEE. Then there's a team of remote AEs, you know, back and back at the factory as it were. But for the large strategic customers, if the phone rings and the FEE says or the sales guy locally says, look, we need an AE and he needs to get on a plane. It's not uncommon at all. Like, you know, for an applications engineer to hear on a Friday that, okay, you need to be in Seoul or Tokyo or California or in San Francisco or in perhaps Seattle on Monday morning. And away you go, you know. Oh, wow. Yeah. I mean, it's but it's just the nature of the business. And part of it is. I mean, you do have two days to get there. So that's nice. Right. I mean, it's not like you have to be here right away. Yeah. I mean, you do hear some stories. Yeah. It happened to me quite often, you know, like depending on the time zones, you know, you want to make the most of your weekend at home. So you would fly at a time where you'd fly into, you know, perhaps somewhere in the Far East and you go straight from the airport to the client, you know, and that, you know, and then that's just one of the one of the aspects of being an AE. But I mean, it can be hugely rewarding. I mean, it can be great because if you work with these strategic clients, you have, you know, you get to work on the leading technology. Like, you know, so if you're doing the best in my case in the past, maybe it was a video DAC or video ADC or audio DAC or audio system on ship or something like that. And then you find them that they're, you know, they're in, you know, like you can imagine, you know, the big phone companies like, you know, I mean, Apple, Samsung. Consumer level. Yeah. Yeah. Something. And then you're there like, oh, wow. Like, you know, when that phone comes out and everybody's just, oh, that's awesome. And then you're going, yeah. Like, I mean, it's amazing. I was involved in the design of that or, you know, the system integration of that. And, you know, like, you really see how many people these big companies take to bring a product to market. Like, you know, and again, in a consultancy role where you're wondering, okay, well, how can I contribute? Or, you know, where you're responsible for handholding a client all the way through to production. If you're dealing with these companies and they're pushing through a million of your units a week, it's a statistical certainty that, you know, any problem in your design is going to surface. You know, so if you have, you know, if your process isn't in control and you're, you know, dealing with silicon and, you know, you've got a, you know, you've got some kind of parameter and it, you know, it's not down to center. Like, you know, your normal distribution just isn't quite tight enough. You know, if you're shipping tens or hundreds of units, you know, you're not one of a problem with temperature or voltage, you'll get away with it. But when you're shipping a million units a week, you gain an appreciation of things like Six Sigma and quality and what the, you know, what the quality engineers are doing and why they're important. And, you know, make sure to be nice to them, like, you know, because they, or, you know, even product engineers, process engineers, you know, all these guys that, you know, you don't typically see it, the value that they add in, you know, in smaller production runs. But once you're shipping large enough volumes, and that goes for startups or SMEs that plan to scale, you might get away with a lot simply because, you know, like, you know, you can just manage it if something goes wrong. Like, I mean, you've a one, you know, you ship 100 units and you have a 1% failure rate. You know, if you yield a 99%, you've got one device coming back and, you know, it's inconvenient to ship the guy another one. And if you're shipping a million units and you've got 1% of them coming back, you have a big, big problem. So, and from an applications engineering perspective, yeah, somebody will pick up the phone and you'll hear about it. And you may be on, you may be on the next flight, you know, you may be on the next plane. And that's just part of the role. But, you know, some guys really like it because you get to see the world and, you know, you get to experience different cultures and everything like, you know.

Kieran Oleary: Sure, sure, sure. So, that's interesting too about the, you know, it always calls back into sharp relief. Like, you know, I'm just a person, I always complain about like, oh, look, I can't get this out of distribution. And that's, you know, as a small person, that's all I really, that's my immediate, that's like, you know, one meter in front of my nose kind of problem. You know, like, it's just what I have to deal with on a daily basis. But then like really thinking about it, like thinking about like what you were saying about how many people are involved in just a single piece of silicon. It really is mind boggling, obviously. And the fact that it's most of the time, it was not even close to develop for someone like me. I'm just benefiting from it. You know, I'm benefiting from the Samsungs and the Apples of the world and the fact that there's, you know, sensors out there that I can maybe use or op amps that were, you know, that Northrop Grumman needed for something. And they, oh, yeah, we'll just sell these to the public too.

Chris Gammell: Yeah, like, I mean, yeah, sometimes, I mean, it can be frustrating. I mean, and I experienced it myself now being on the other side, you know, dealing with much smaller clients. But sometimes you don't want the cutting edge silicon, like, you know, for, you know, any number of reasons. But, you know, if you think about these things from a business perspective, if something goes wrong with new silicon and there's a workaround to be had or an issue to be investigated, the strategic customers are going to get that support. They're going to have an applications engineer on the plane. They're going to have the local FE working the long hours. They're going to have their sales engineer or account manager, you know, speaking up for them within the semiconductor or semiconductor companies organization. Like, but if you get that cutting edge silicon, who's going to speak up for you, you know, if something goes wrong? So, I mean, when I was in Xilinx, actually, as a graduate, somebody said one time, and it's always stuck with me that you want to be on the leading edge, but you don't want to be on the bleeding edge. And because if you're a small owner, and the Xilinx felt that they had that problem themselves. Like, I mean, at the time, they were quite frustrated because, you know, FPGA is kind of always push the process node. Sure, yeah, seven nanometers, right? Yeah, and I'm sure guys in Intel or, you know, AMD or whatever will tell you the same. They feel like that they're doing all the research, you know, or IBM or whoever. They're doing all the research, and other companies are benefiting from it, you know, from it. So, if you're an organization that can manage those relationships and absorb the cost if something goes wrong, you deserve the benefit of being on right at the edge. But if you're a small organization, maybe bear in mind that, you know, can I do what I need to achieve from a product perspective without getting the latest whiz-bang thing? And sometimes you can't. Sometimes, you know, your product is a hardware product that needs absolutely, you know, the cutting-edge stuff. But sometimes you don't. And, you know, like, I can completely understand that it's frustrating for smaller customers. But, you know, it is a business. And, you know, the guys with the most money, you know, they're going to get first dibs on support and access to silicon and everything. And, you know, sometimes it's good. Even picking your suppliers. You know, there's some companies that I typically won't design in almost at any price. So, you kind of know who I'm talking about. I don't know of any of these that, yeah. Like, I mean, you know, you can guess.

Kieran Oleary: Here's the thing I always think about them. And if people have listened, yeah, exactly. People have listened, they've heard the show before. They know who we're probably talking about. The ones who get that tier one support probably freaking love them, though. Oh, I'd imagine so. It's just like how they're set up. Their setup's so different. And it's fine, you know, like, but it's just, you know, there's some companies, like you're saying, do not support the smaller, mid-sized enterprises.

Chris Gammell: And I think that's perfectly fine. I think the problem comes, and I've worked in companies who've had this discussion. You set out your stall. You set up the business a particular way. And if your business is targeting tier one companies in automotive or tier one companies in consumer, and you, you know, like you pursue that market. And then you just tell, you know, even tier two companies, let alone tier 10, if you like, we're not interested in your business. You know, we don't think you're bad guys, but it just doesn't suit our business model. Yeah, that's fine. And then you get other guys like TI who support everybody, and they've got their application structure and their sales structure and everything set up to do that, to support everybody as best they can. And that's fine. The problem is the companies that kind of maybe get a bit greedy. Well, let's, you know, let's manage the relationship with the tier ones, and then let's just sell to the, you know, tier twos, tier three. Right, we'll sell the offcuts, you know. And then you're going, well. You can have the soup bones, you know, the tripe of our... Yeah, and you do wonder, for the amount of money that they make out of it, you know, revenue-wise, is it worth the reputational damage or, you know, people going, oh, they're just difficult to work with.

Kieran Oleary: I don't think the amp hour is going to be any kind of dense, unfortunately.

Chris Gammell: Yeah, like, so, yeah, it's interesting. And like I say, applications engineers, particularly if they keep their ears open, you know, because they're hanging around with the sales guys and that sometimes, maybe get exposed to these kinds of discussions earlier in their career than a design engineer would back in the factory. Because when you start out as a graduate all the way maybe up to senior design engineer, you're expected to make absolutely the best performing, you know, circuit design that you can. And that takes a lot of time. That's a full-time job in itself. And as you become more senior in the design organization, that's where the discussions on price and perhaps other things come in.

Kieran Oleary: So what about, were you making, like, dev boards then for customers, like ones that would be broadly distributed? Oh, yeah. Like ones that, like, we would buy off the shelf? Because I'm always curious about that process, like, internally on chip companies.

Chris Gammell: Again, it depends on, you know, on how your company is set up. But, you know, a lot of the evaluation boards, development kits, reference designs and so forth that I worked on would never be seen on DigiKey or Mouse or Farnell. Because, you know, if the silicon isn't made for the general market or mass market, then there's no point in releasing the development kit to the mass market. But, yeah, some of the boards I worked on would be available on these, you know, distributor websites to support the silicon. But, of course, at a horrendous markup. And some of that is the price of the components. But some of it is typically a strategic decision that perhaps you might see a board for $500 or $1,000. But, of course, you know, well, you know, you'll know, like, that it doesn't cost $1,000. It might cost $100, $200. And they're not trying to make money out of the boards. They're trying to discourage, you know, what you might call the tire kickers, you know, like where somebody wants to do a hobby project. And my background in audio electronics, you see this a lot, that you get guys that are really enthusiastic about audio. And then they get really frustrated. Oh, why do I have to pay a couple of hundred dollars for an audio DAC reference board? You know, it's a very simple board. It's got, you know, some pen header, you know, interface to the DAC and some jumpers to control the settings. Why is it so expensive? And the reason is, you know, kind of frankly speaking, that, you know, the semiconductor companies are in business. And if there's any issues encountered with the, you know, with the guys buying these boards, perhaps they'll never recover the cost of even supporting the customer. What would be the interactions at the beginning?

Kieran Oleary: Is it just like, well, let's just come up with something interesting? Is it an eval board that it's just like, well, let's just break out the pins and make it do a thing? Or is it like, let's make this do an interesting thing in the case of like more commercially available dev kits?

Chris Gammell: It really depends on what you're targeting. It's not uncommon to see three or four different types of boards. And I mean, I have no commercial relationship with Nordic Semiconductor, for example, but they're a company that seems to do this well. You know, you might see like, OK, you've got a very small, you know, relatively speaking form factor board that will show one of their Bluetooth devices, you know, on the size of a coin cell or something like that. And then they'll have, you know, a larger board, which you might consider, you know, a reference design or something like that. You know, which is, you know, a lot more broken out, but allows you to, you know, to do a little bit more. And then they'll have a full scale development kit that will allow you to do current measurements and things like that. And I think they're a good example of a company that does it well for the mass market. And like I say, I've got no relationship with them other than I think that they handle that kind of, you know, side of things well. And they are clearly targeting mass market with the IoT. And, you know, somebody has sat down there and said, well, we need, you know, a form factor thing. We need something that people, you know, the former engineers can get their hands on. And then we need something that, you know, things like current measurements are actually quite important, like for pretty much for all products nowadays, but certainly for anything battery powered. And you need, you do need a different board. You do need a different board for, you know, for these different tasks. Of course, internally then, which the customer will never see unless perhaps you're a tier one customer and you're given one out of courtesy, there will be a whole lot of range of boards supporting the device as well. So, you know, you'll have design evaluation boards, which may be, you know, a much grander version of the board that a customer might use for customer, excuse me, for current measurement. The design evaluation or design verification engineers may have, you know, a much more sophisticated version of that board that can be used for, you know, lots of performance related measurements, you know, like my background in the audio space, you know, for taking measurements, I guess, in R and THD, current measurements, of course, these kind of things. The board performance needs to be very, very high because you don't want to be measuring the performance of the board. You want to be measuring the performance of the silicon. And then sometimes if you want very, very high performance, you know, on an analog power supply, you might need to, you know, spend a little bit more money or take a little bit more area in the power supply section of the board, which, of course, you don't want to pass along to the customer if they just want to evaluate the part with all the supplies ganged up. So, and then, of course, then there's internal test boards, which can be, you know, quite complex boards then. It's endless depending on the type of silicon that you have or the device complexity. Got it.

Kieran Oleary: Yeah. So what would you usually end up getting? I mean, would you start with something that was like hand-bonded silicon that's like basically just an eval, like coming out of the fab? Or like, what was that? So like, let's walk through maybe something's, a new product rolling off the line. Can you walk us through what the process would be like?

Chris Gammell: From a design evaluation or applications engineering perspective?

Kieran Oleary: All the way. So yeah. So we're going to say the new chip, awesome DAC 45, right? Awesome DAC 45. Awesome DAC 450. That's even better. Better than 45 is awesome DAC 450. So you have been talking to the silicon designers and the production folks. And you're like, all right, we're getting ready to go. What happened with like, what were the boards that you would make? And then how would you then interface with the customer in order to work with those? Like, how would they get those boards and work with them?

Chris Gammell: Okay. So give you the nice fairy tale or the reality.

Kieran Oleary: So, I mean, let's have a fairy tale with some cautionary, you know, big bad wolves in there. Okay.

Chris Gammell: Well, I mean, I can give you, okay, I guess the big bad wolf is that the silicon comes back and that there, you know, there's a fatal flaw and that, you know, the thing won't boot or it's even happened. You know, it's shorting out or something like that. And then. It's a crib death for silicon. Yeah. And that does happen. There's some kind of really funny stories that you can tell about that, but I won't embarrass anybody. But, so assuming that the silicon comes back and it's in decent shape, it's an engineering sample. Then if you're dealing, you know, like with, say, in the consumer electronics field, you know, timelines are very, very compressed. I mean, and you can imagine you want a new phone every year or something like that. Timelines are compressed. So it literally can be that after the, you know, the silicon is, you know, comes out of the fab, it goes, it gets assembled, you know, it just puts in a package, you know, see if the silicon die gets diced up, you know, off the wafer, gets put in a package and assembled. No time for testing. You know, maybe you have a couple of hundred and a couple of thousand devices on a reel and there's boards waiting for you. Now, typically you won't, it does again depend on the device, but quite often you will have, like I said, an evaluation board, but you may have a socketed version of the evaluation board where you'll do two versions of the evaluation board for absolute performance. You'll take your silicon or your package and you'll pop it down and you'll, you know, reflow or rework the board to take the silicon. The board is already done. All that's left to be, is to be put on is the device and you'll have a technician do that for you.

Kieran Oleary: Okay. But before we go further then, so when did you get the spec internally that allows you to make that board?

Chris Gammell: Oh, pretty much. I mean, you do like, I mean, if you have a good applications engineering and design evaluation team, you will know before, you know, before silicon goes to the fab, you know, your applications engineers and even your PCB layout engineers. If you have those internally, they, they will be working, you know, with the silicon designers to know what the ball out or the pin out is before the silicon even goes to the fab. And the reason that's important, and I've been involved in projects personally, where you have a CSP or a BGA and the designers do the IC layout or, you know, you've got a, just like with a PCB, you've got PCB layout for an IC. You have IC layout and there's guys that specialize in that and they, they lay out everything. They think it's fantastic. And then it comes up with a pin out, which the PCB layout guy looks like and looks at and goes, oh, I simply cannot route that out, you know, so back to the drawing board. And then, you know, and, and it does happen. It made, it, it, it literally made the cost.

Kieran Oleary: Oh, I, I was amazed. I was amazed when a early, I had some access to some, some early stuff and they said, you know, look, this is not a finalized pin out yet. And I'm just like, what, what do you mean? It's not finalized yet. It's like, yeah, it just might move. Yeah. What are you talking about? You know, and the, okay, I guess I won't design anything to your pin out. Yeah.

Chris Gammell: But geez, like it does happen certainly. And it, I mean, again, depending how quickly things are moving, you know, the silicon might go, you mightn't even have the pin out finalized. And okay, well, when we're, you know, putting, you know, we're bonding it out, you know, that'll come later. And, and that does happen with, you know, with very compressed timelines in, in, in, in certain industries. Of course, you want to avoid it if you can, you know, for quality and to, to reduce the risks of making mistakes, but, but sometimes you can't help it. So you want your PCB layout guys involved. You want your applications engineers guys involved. And as soon as the silicon goes, probably even before it, your, your, your guys will be doing the schematic capture for that pin, you know, for that PCB. And then, you know, just like any other PCB, you'll review it, hopefully do a schematic review, do your layout review. And then, you know, your, your PCB, you know, typically, of course, your PCB will come back quicker than the silicon because, you know, it just takes longer for silicon to. For now, for now, man. Yeah, the chip printer. Yeah. Yeah, right. You've been listening to a lot. Yeah. Yeah. Yeah. So I can't wait till the chip printer comes, you know, the. Yeah. Right. So it, so that all happens. It comes back, you know, and like I say, you might have a socketed version where, you know, like these, these chips are coming back and they won't have been tested. You won't have time to test them quite often, or, you know, you won't have a test program in place. And I see test program in place because the test engineer won't have time to, you know, they need physical samples to test their test program before they can deploy it, you know, to the assembly site, perhaps where it will be tested. And, you know, so quite often people will design a custom socket and they can be quite expensive. I mean, they can, those sockets can run to thousands of dollars, you know.

Kieran Oleary: I believe it. I mean, especially if it's like BGA or something, it's got to be super perfectly aligned and like connections out to everything.

Chris Gammell: And, you know, you don't want to lose performance. Now, like I've worked a lot, you know, in the audio space and, you know, where, you know, you might have a THD measurement, you know, so total harmonic distortion. So without knowing too much about audio, you can imagine distortion is bad. And if you're giving up some performance on the connection between the socket and the device, your THD measurements can look quite poor because you don't have a good, you know, electrical connection between the socket and the device. And if you're pushing the boundaries of a given technology or, you know, for any number of reasons, your performance may not look fantastic in the socket. So you might take, okay, well, you know, get somebody to test 10, 20, 50 of these boards in the socket, you know, for functional, you know, functional certainty. You know, excuse me, certainty that they're functionally operating or at least, you know, some way reasonably functional. And then you would take some of those devices and have a technician put them down to the evaluation version of the board that's, you know, done, you know, to, you know, to have the silicon, you know, placed on the PCB. And that generally happens. And if you have time, you will do it at your home base, you know, back at the factory and you'll take as long as you want. You'll evaluate all the parameters and so forth. And then when you're ready, your FEE, perhaps a local guy, you know, an applications engineer or a marketing guy and a sales guy will all go happy out into the customer and show off, you know, DAC 450 or whatever you said. But it does happen where those timelines can be extremely compressed, where you will have silicon will come in untested on Thursday or Friday. And, you know, you'll test them in a socket, you'll get the 10 best golden devices, you'll pop them down on some evaluation boards or, you know, demo boards for the client. And then on Sunday night, you're on a boat, you know, on a plane to, you know, the West Coast of the US or somewhere in Asia or something like that. So that early the next week, you can bring in and hand over five or 10 demo boards with working silicon to a client. Now, you want to avoid that if you can, obviously, because the more time you have, the better, but the world doesn't work like that. And sometimes you just need to, you know, to have compressed timelines. It's even happened on a couple of occasions when I was in Seoul in South Korea. And I was sent to Korea to get everything ready before the silicon was available. And the silicon followed me to Korea and I had the evaluation boards waiting with me in Korea. And we we got a local, you know, a local company to to rework the boards and put the silicon onto the evaluation boards so that we could bring them down to a large, a large Korean client, which maybe you work out who they are. But, you know, you know, sometimes, you know, sometimes when you're working consumer electronics, it can be that compressed. And if your competitors are doing it, you know, like you have to do to, you know, even if even if you do wonder sometimes, is it really this critical? You talk to the engineer and the client then a week later. Oh, yeah. Thanks for that. It's sitting on my desk. You know, it happens.

Kieran Oleary: So I'd love to move on to the consulting stuff in a second here. But how is that different then? So that's an evaluation board. How would that then change if you were going to be sending out a board to the general public? Is that more like future revisions of silicon?

Chris Gammell: Perhaps, but it might be that the the expectation for a board for the general public would have higher, you know, higher requirements in terms of the support and collateral. You know, so you might want to have, you know, make sure that you've got a really good, clean schematic that you'd be happy to show to a client, you know, or, you know, a customer. You know, you might want to even make the Gerbers available, you know, quite often, you know, for a silicon company, you know, they'll make everything available. They'll make the Gerbers available. I mean, it doesn't cost them anything if you copy the board. But, you know, they might want to, you know, make it easy to see what the layout under a BGA or CSP or something is like. And then, of course, you know, they'll give it to you in a PDF. They might even give you the original files, you know, Altium, Eagle, whatever it is, all this kind of stuff. Yeah, how about KiCad? Come on, man. Yeah, KiCad. Well, again, it does depend, you know, a lot of the silicon companies, you know, they've got actually vendor relationships with, you know, with maybe companies like Mentor Graphics or Cadence or something. And they might do an overall, you know, they'll do an overall package by, you know, for their various tools. So, you know, it might be some time before you see KiCad. But, again, if you're targeting the mass market, I'd imagine, for example, IoT, you know, and it would make a lot of sense to support things like Eagle and KiCad and so forth. Because there's a lot of, you know, perfectly profitable companies that will buy plenty of your chips that maybe don't need Altium or, you know, they don't need DX Designer or something or pads or something from Mentor Graphics. And it would be, you know, they'll be able to give you money, you know, and if you can help them out. Right, exactly. Right.

Kieran Oleary: Money still talks. Yeah, absolutely. You know, using open source software in order to help promote the product. Absolutely.

Chris Gammell: And it does come back to, you know, what your business model is. If you're targeting tier ones and the tier one uses Altium, given the Altium files, if you're doing… That's right, exactly. If you're doing… Service business. Yeah. If you're doing horizontal, yeah, then you have to look at what the market is doing. And some of the companies that you would, you know, consider good at supporting, you know, a broad base of the industry, you know, they can be quite, you know, quite supportive of platforms like KeyCAD. But, of course, again, they are a business and, you know, they need to, you know, they need to manage the time that they put into creating, you know. Sure, sure, sure.

Kieran Oleary: I know. Mine was mostly just a general gripe, not an actual thing. Oh, yeah. I don't expect it any time. But it's fair.

Chris Gammell: But it's fair. You know, I think some companies need to do it, other companies don't, you know, depending on what they're targeting or maybe should do it rather than need to do it.

Kieran Oleary: So, on the consulting side, so then you transitioned into consulting. So, what does that look like? I mean, does that look similar to the… I mean, you said that some of the stuff had prepared you for the consulting side of things, but what does that look like on a daily basis now? Yeah.

Chris Gammell: So, yeah. So, I was, I think it was 2012, I went out my own. I left Wolfson Microelectronics and I started doing, you know, my own consultancy work in, you know, Mixed Signal Systems Limited. And there's lots of reasons for becoming a consultant. So, a kind of, I guess, a lot of guys down through the years have done things to impress a girl. And a lot of the reason I became a consultant was for that reason. Like, so, my now wife worked in Wolfson Microelectronics as well. And I'd always said, you know, like, oh, I'd like to, I want to go to my own. I want to do my own thing. I'd like to be my own boss, you know, this kind of thing. And, yeah, she called me on it and said, well, you keep talking about it, just do it. So, yeah. So, that day happened to be the day for my… Oh, man. Yeah. It happened to be the day for my performance review. So, I went then to my manager, a good guy, and he was getting all the paperwork ready for, you know, your performance review or whatever. And I said, I'll save you the paperwork. I'm off, you know. He said, what? I said, yeah, I'm going to go to my own. Oh, man. Yeah. Wow. Yeah. So, one of it, you know, I've kind of been a bit flippant. But, you know, but, yeah, she did really call me on it and said, well, like, what are you waiting for? If you're going to do it, do it. And, yeah, I'm glad she did because, you know, it's worked out. But the other reason is I was traveling really, really a lot as an applications engineer. And maybe I didn't want to travel so much.

Speaker ?: Right.

Kieran Oleary: You just say you were in Korea quite a bit. And that'll take a toll on the body. Yeah.

Chris Gammell: Like, yeah. I mean, if anybody has worked with the Koreans, I mean, did they really know how to work hard? Like, you know, I really enjoyed my time. And party hard. Don't forget that. Don't forget the soju. Yeah. Yeah. The Irish of the Orient. Yeah. That's great. Yeah. I really, I really enjoyed my time there. I really liked it. But I was never based full time in Korea. But at the end of my time in Wolfson, I was there for five years. I totted up my time. I was five years in Wolfson and I was a year and a half all in in Korea. And I was never based in Korea. Oh, my goodness. I was out of 25% of my time. And that was tough, you know, after a while. You know, like, so, yeah. Anybody gets the chance to go to Korea, you should go there. It's a wonderful place. Oh, man.

Kieran Oleary: So I used to work for Samsung and I, like, my coworkers were amazing. But, like, so, such a hardworking, like, top-down, like, militaristic style management. And it's just like, oh, I did not do well there. And, like, that's almost like the opposite of what consulting is, which is kind of interesting. It's like, instead of, like, being told exactly what to do by your superior who was told by their superior, by told by their superior. Now you're like, oh, I guess I'll just work for this one person. It is.

Chris Gammell: Yeah. It is quite a change. But, you know, I don't regret doing it. And nowadays I have two types of clients, really. I have SMEs and, you know, startups that will hire me to do product development work, you know, in the mixed signal space. So, you know, they have an IoT product that they have in mind. So, you know, a lot of sensors and maybe a lot of analog design, some RF design, you know, firmware development. Typical product development that you might expect from an EE, sorry, excuse me, working as a consultant. I don't typically have medium-sized enterprises as clients, typically, because if they wanted a hardware engineer, a firmware engineer, an RF guy, they would hire those people. So, but, and I think a lot of consultants would kind of say the same. But then with my background in semiconductors, I do some consultancy work for the large semiconductor companies, whether that is, you know, how to engage with clients, you know, like, so perhaps how to engage with clients from the Far East, you know, from Korea or China, that kind of thing, because, you know, there is cultural differences and so forth. And like you say, with Samsung, yeah, there is an expectation of how work is to be done. So some of that type of consultancy. And then, of course, technical consultancy as well. So maybe there's a particular block of work they want to outsource, like, you know, write a data sheet or design an aval board or a reference design or even help with some aspects of IC design. I don't typically do analog IC design. You know, I do perhaps some digital IC design because, you know, from my background in FPGAs, it's largely similar. But there may be problems with performance of a particular IC, and there might need to be design evaluation or design debug, you know, around the signal integrity space or, you know, analog performance space where there might be noise coupling or this kind of thing. And that's the kind of work that perhaps would suit a consultant for a large company. You know, you don't need somebody full time for that. But the issue nonetheless does need to be resolved. And they either need a bit of help because they've kind of struck out trying to do it themselves or they simply don't have the bandwidth. You know, it's quite common. And they just need some extra help.

Kieran Oleary: Yeah. Does that mean you get pulled in like as almost like a contractor more than a consultant then? I mean, in that case, because it sounds like I just wonder about the integration with a large scale company like that. Because you're basically, you're acting kind of on behalf of them instead of just like being like delivering, hey, I'm going to make you a board. It's not like, hey, I'm going to, I guess writing a data sheet, you could deliver the data sheet. But some of that stuff sounded like it was more of a being embedded in the company more.

Chris Gammell: Yeah, sometimes. Yeah. Like, I mean, there is, and sometimes it's nice for me, you know. So I have a home office and, you know, you'd hear other consultants say as well, you know, it can get quite lonely if you're doing that all the time. And sometimes it's nice to be on site, you know, with, you know, with other people. And quite often you have to be, you know, if you're debugging silicon or something like, you know, I have quite a good setup at home.

Kieran Oleary: I'm just going to take this home.

Chris Gammell: We're cool. We're cool. Yeah. And I'm just going to bring your spectrum analyzer with me. And, you know, like if you're working the audio space, you know, like typically you have an AP, an audio precision analyzer. And you go, yeah, I'll just take it on the bus. It's fine. You know, of course, that's not going to happen. Like, you know. So, yeah, sometimes you need to be on site if you're doing a debug or something like that. But, of course, writing a data sheet you can do from anywhere or if you're doing a reference design, quite often they're happy for you to, you know, like a lot of the work with a reference design or something. It is a block of work that can be all encompassing and can be just given to somebody to do. And sometimes they want it done on site and sometimes they don't.

Kieran Oleary: You know, it's like a mini product almost. You know, it's like the output is basically a board and, you know, the stuff surrounding it.

Chris Gammell: And, yeah, there is and with a background in, you know, doing this, there is a little bit of a difference designing an application, you know, a reference design or an evaluation board because it is a deliverable that a client will see. And if, you know, you know yourself, like, but quite often maybe schematics, like there might be a spelling mistake in the schematic. And in some sense, who cares? But in another sense, this is a deliverable that a client, you know, they're going to see. And if, you know, if you're making spelling mistakes or, you know, calculation mistakes and notes on a schematic, well, I can't see what's inside your IC, but I know that you didn't take the care to, you know, to do the schematic. Well, right. You know, and so there is, you know. It's like a best foot forward kind of thing. Yeah, it is like the Alicia from, you know, Embedded, the Embedded podcast, you know, she speaks, you know, a lot on Jack Gansel and some other guys about the concept of technical debt. And there is a thing like, you know, you get the small little mistake there on the schematic, but that stays with you all the way through every client that has that. Like, I remember one, I'll never forget it. We had a stabilization capacitor, you know, for an internal LDO on an IC. And in early version of the schematic, it was, or the data sheet, it was down as one microfarad and it needed to be 4.7 microfarads. And every project in the large client did it at one microfarad. And most of them, the FEs caught or me and my colleagues caught, but one or two got through. And then when they were doing their early production ones, when these large clients do an early production, you know, there could be 10,000 units, you know. And it's like, oh, yeah, it's just noisy as heck. And what's going on? And then you look at the schematic and like, oh, like. And so, like, even though, again, with small clients, maybe you don't see this thing in early production. It could be 10 parts, but when you've 1,000 or 10,000 parts, there is an expectation that you get this right and quality is important. So sometimes working with these companies, if they don't have the resource available internally, it's nice about to give it to somebody who has an appreciation of what can go wrong if you don't pay attention to, you know, even small design notes in a schematic or in a data sheet.

Kieran Oleary: Well, you had mentioned when we were talking before the show, too, I mean, you talk about doing signal integrity stuff, and that's kind of part of the mixed signal system side of things. What are some of the common mistakes you see either clients, not even clients doing, but what are some of the things that you feel tripped up by or you think listeners might be tripped up by in the mixed signal space that they should be watching?

Chris Gammell: Okay, there's one and one only. And it's, I actually have it written on my schematics, on all my schematics that I do. It's down as a note at the bottom of, as part of the frame. And it just says, current flows and loops always. Either you or James Clerk Maxwell is going to choose the return path. And that's it. If you remember that, whether you're doing digital, RF, analog, it doesn't matter. If you remember that there is a return current, and just think, I don't care how complex your design is, just remember the light bulb. You know, once you flick that switch on the light bulb, current is coming back. And it is something, you know, and Howard Johnson speaks about this, you know, that high-speed digital design guy, the author of that book. You know, he mentioned that perhaps... Ask us the show before you return. Absolutely, yeah, absolutely. Great episode. My favorite episode, actually. The, yeah, and he's a good guy, actually. He's a great guy. But the, you know, he speaks to the fact that, you know, maybe, particularly digital designers, you know, you put, you know, you put a ground symbol on your schematic and that's it. Like, you know, you're done. And that probably accounts for, you know, and that accounts for, you know, a lot. Because that ground symbol is where, you know, where good signals go when they die. And, of course...

Kieran Oleary: It's like a bucket, and then eventually there's someone that kind of carries that bucket back to the source, and then it's fine.

Chris Gammell: Yeah, it's... And that really is key. If you... A lack of appreciation of the return current, I would suggest, is the root cause of... Well, maybe it's not the root cause, because, of course, there's physical root causes. But it's certainly a strong contributory factor to the majority of signal integrity and quite often EMC issues. Because you hear people talk about things, oh, I need a split ground plane, which, you know, like one of those religious wars, you know, whether you do or you don't, you know. But a lot of it comes down to, well, I need to split ground plane. And essentially what you're trying to do is manage the return currents. But if you understand why there's a return current, the properties of the return current, then you can, you know, for very, very high performance, DAX and ADCs or PLLs is another one, you know, phase lock loops, are ones that can be quite sensitive to noise on the ground plane or even the power plane. An appreciation of why, you know, why you have a return current, remember the light bulb, and how that return current is going to behave on a power plane or on a ground plane. That is more effort in understanding what's going on there would solve really a lot of problems. There are secondary issues as well, but by an overwhelming majority, I would suggest that that's something if people understood, they would, you know, they'd probably make a lot of progress in terms of their designs, you know, simple DAC or ADC all the way up to the most complex systems like, you know, the RFSOC from Zylinks that we spoke about earlier.

Kieran Oleary: So let's talk about the DAC 450, you know, DAC 450 has just came out. It's a great new product here. A really good development board, I heard. Was that what we were calling it? DAC 450? DAC 450, yeah. New DAC 450. Yeah. I got rebranded anyway from the marketing guys. Exactly, of course. Of course, you know, it didn't test well in the marketing group.

Chris Gammell: Yeah, 45 didn't work well, so 450 sounds better. Yeah, right.

Kieran Oleary: 450 really brings more power and, you know, more of a, it's a bigger number. As some say. Yeah. Anyways, so DAC 450, you just designed a new design. What is, so it's failing emissions testing, I guess? Is that kind of the thing that you would normally see? Like, what are some of the problems you would see? And then how would you diagnose it back to the point where you said, hey, current's flowing loops, and I just found this, and I'm pointing at this on your schematic or your layout? Yeah.

Chris Gammell: Okay, well, it is, and there's two ways of looking at this, or two groups of people might look at this. So the design engineer who designed the product, and he's, you know, he brings it, you know, for pre-compliance testing, or, you know, if they have a chamber in-house, or, you know, he goes to the, you know, to the external house to, you know, lab to have a test, and he gets a report back. And, of course, the design engineer knows his product quite well. If you're an applications engineer, of course, supporting the, you know, the DAC 450, you perhaps have more limited information, or certainly at the beginning, you have less exposure to the device. So, or, excuse me, to the product and the layout and so forth. But first thing, yeah, you, like, always, you know, look at the schematic, you know, like, particularly if you're responsible for, you know, for that device, you look at the schematic and make sure that the implementation is correct. Because you will see things like, I mean, how that problem I described earlier, as it happens, came up, you know, it was a stabilization load capacitor for the LDO, and it was spewing out all kinds of noise, because, you know, it was oscillating. You know, you will see that, you know, if you're doing emissions testing, you know, a near field probe, you might say emissions testing, of course, it sounds very formal and everything. But, you know, you'll get energy radiated, electromagnetic energy radiated, you know, in the near field, and you might recognize it as an electric field or magnetic field coming off the board. And quite often, start at the schematic, make sure that the basics are right. You know, and it sounds silly. So, you know, why would you do that? But quite often, you'll capture quite a lot, or you'll get a feel for the design, if perhaps the power supply looks kind of a bit open.

Kieran Oleary: Does it look exactly like the reference circuit, or is it, you know, is there something around?

Chris Gammell: Yeah, it's amazing, God, like the kind of variety of things that you can see, like, for something, even for a simple device, you go, like, you think, you know, there's a phrase that I heard, actually from a guy in analog devices when I was there, and said, like, there's nothing foolproof to a sufficiently talented fool. And it's really amazing what people can do. But to be fair to people, you know, some of these devices are quite complex, and you have to give them credit, like, you know, that maybe it just takes a bit of hand-holding. So that's the first step. Looking at the power supplies are always important. Quite often, you'll get a lot of noise, either conducted or radiated, you know, from the power supplies. But assuming that everybody is done, you know, at a schematic level, that everything's in kind of reasonable order, you will start to look, you know, perhaps at the layout, you know. But before you do that, you might take a near-field probe, because I'm thinking of, well, what do most people have available to them? And, you know, not everybody has a big, you know, like a big chamber that they can just do RF testing in. Like, you might want to just, you know, take your scope probe, you know, I think maybe Dave Jones might have even done a video with this, and, you know, short out the ground probe, and, you know, short out the ground, you know, the ground and the probe tip. And you can make a little H-field probe, and you can run it over your board. And you'll quite often see, you know, around the clock or something, say, perhaps you have an audio DAC or something. You might have a clock there at 12.288 megahertz. And you might see that or multiples of that, you know, turning up somewhere on the board. Now, if you see it over the audio DAC, then maybe that's okay, because the clock is going into the DAC. And if you see it near a crystal or an oscillator, that would be fine too, perhaps, of course, depending on the level. But one of the things that really causes a problem is, again, current flows in loops, you know, like we said, think of the return current. And if the signal trace passes over a split, you know, so, for example, you've got a four-layer board. Your signal, the clock signal is on layer four. And on layer three, you have the different power planes, 3.3 volts, 1.8 volts, and so forth. If your clock signal passes over a split, you know, on layer three, which you have to because, you know, 3.3 volts and 1.8 volts can't be shorted, you will, you know, cause quite an effective antenna. I mean, I mean, and it's quite remarkable, the amount of energy that even one signal, digital signal, traveling over a split on an adjacent layer can cause. And that's one of the first things I would always look for, you know, as an applications engineer. If you're okay, well, your schematic looks fine. You know, when we've looked around on the board and we say, okay, God, there's just energy oozing out of this section of the board here. And it doesn't really seem to make a lot of sense. Let's look at your layout. And if as soon as you see signals crossing splits, then you're going, you know, like, here's a real problem. And in a more general sense, perhaps, if there was one tip I could give to startup companies doing hardware products, it would be to consider this kind of problem quite early. Because unlike a software company where if you have to, if you found it, you know, okay, we've made a big error, you know, and we have to do a software patch, bar you've made some terrible architectural, you know, decision, that means you can't scale up your product and the product won't scale. Generally, you can fix a lot of problems by just releasing, you know, releasing a software patch. The same isn't true in hardware. If you find you've got a problem like that, perhaps there's nothing that, you know, an EMC engineer or signal integrity engineer can do to help you. Now, it's possible that you can and perhaps you could shield the whole product. But if you have a Bluetooth product or a Wi-Fi product with a plastic enclosure, then of course you can't put a metal shield over everything. So you're not able to just enclose it and forget about it. You might have to go back, do the PCV.

Kieran Oleary: Some of the stuff you actually want to be emitting from that product is just in a very controlled way, hopefully.

Chris Gammell: Yeah, of course. But then you've got this garbage, you know, that's being emitted as well. And, you know, you're failing your compliance testing, CE marking in Europe or FCC in the US, you know, or whatever. And then, you know, you're really well-meaning. External test house tells you, look, guys, we can't sign this off. And often the test houses are extremely helpful. I know you've had guests in the past that, you know, talk about this, you know, but they can be very helpful. Like, and, you know, there's some things like, okay, you put down on key traces to connectors coming off the board. Maybe you don't think you'll have a problem, but you'll put down an unpopulated, you know, 10-puff capacitor to ground. Or you'll have a zero-ohm resistor that can be replaced with, you know, a wee inductor or a ferrite bead or something if needs be. And if you put those hooks on the board, the test house quite often, you know, they know what they're doing. They'll go, oh, look, you've got a problem radiating from this connector here. You know, but I see that you've made our job easy. You've put in some zero-ohm resistors there. Well, we'll just, you know, take those off and put in a wee ferrite bead or some type of choke or something in there to help you out. And we'll get the job done. But you make their job much easier if you accommodate, you know, those kind of things on the boards. But sometimes if you have a signal running over a split, there's nothing that can be done, unfortunately. And if you're, like, you find it kind of comes up, you know, like with hardware kickstarters and that. And the guy says, you're going to have to redo those PCBs. And they go, oh, gosh, we don't have the capital to do that. This is our product. We're shipping it. Of course, yeah. And that comes up a lot, you know, dealing with different types of clients. But, and it's really disheartening, like, you know, and quite sad. Like, you know, because these people, you know, poured their heart and soul into this product. But they just, they weren't aware that this was an issue, you know, that they would need to address. And then, of course, because it's quite late in the day, when it arises, it can be difficult to solve. And even if it can be solved, then, you know, it is quite often, it's great for consultants. They ring up in a panic. Oh, my God, we can't ship our products because we can't get our paperwork in order. We'll pretty much pay you whatever you want. But this has to work. And you're like, okay, great. Like, you know, fantastic. I'll get, you know, I'll get my near-field probes. And quite often, you can help people out, you know, particularly, you know, look, okay, you know, you have to make those slits smaller. Like, and there might be a trade-off in thermal performance, you know, but perhaps you need to make the slits a little bit smaller. Like, you know, so that, you know, a given wavelength doesn't find its way easily through a slit in an enclosure or something like that. You know, and that can be easily done. Copper taper, you know, you 3D print, you know, well, maybe perhaps by the time you get to production, you wouldn't be 3D printing. But, you know, there's things that can be done. Whereas, unfortunately, in some products there can't be, you know, you have to redo a spin. And, yeah, so if you are a startup, you don't need to understand all or even anything, you know, that I've been speaking about in the last couple of minutes. But just remember, allow yourself some budget to do a re-spend. Factor it into your schedule and your costs. And then bear in mind that this is, you know, put it on your project plan. We need to do, you know, compliance testing or regulatory testing. Call it what you want. You might need, you might not understand what needs to be done, but there's plenty of people that can help you. But if you don't put it on your project plan, you'll just forget about it and you might find yourself coming unstuck, you know, unfairly. Or, well, not unfairly, but, of course, unfortunately.

Kieran Oleary: So back to the integrity side of things, you're saying a lot of the things you might see are, you know, just jumping split planes and stuff like that. Yeah, like or. Other common issues you might see?

Chris Gammell: Termination resistors in the wrong place, actually. You know, like, and this is a mistake. I mean, I've made all these mistakes. Like, I mean, I could write a book on the mistakes I've made, both from a business perspective, you know, because everybody learns, you know, when you be a consultant, but from a technical perspective as well. But signals running over splits, termination resistors in the wrong place. You know, that's quite a common one as well, you know, because, you know, like I said, when I was, you know, in Xilinx or in ADI and the video DACs and video ADCs, you know, a video DAC or ADC or a HDMI transceiver. You know, the RGB lines, you know, there could be 36 bits wide coming out of an RGB ADC, you know, plus the clocks and the sync signals. So RGB, you know, each of them 12 bits and could be 36 bit bus plus clock and the syncs. And, you know, you know, if you don't get the termination resistors in the right place, then you can have a whole bunch of energy, you know, bouncing around that board, you know, as it tries to dissipate, you know, it runs down, you know, it runs down an unterminated line and it comes back and it just keeps on reflecting until that energy is dissipated. And that can be, you know, like, you know, and if that's on your clock signal, that's quite problematic. Like, I mean, perhaps a way to describe it, you know, like on a podcast might be a clock is supposed to go tick tock, tick tock. Whereas if you have an unterminated line, you might end up with tick tock, you know, and it's like it's really not what you're looking for, because if that happens, then, you know, you're sampling at the wrong time. And aside from the EMC issues that you will cause is that energy needs to dissipate somewhere. And, you know, you'll say, OK, when it's a series termination resistor, perhaps think of it as a source termination resistor so that you put it at the source of the signal. And end termination resistor goes at the end of the line. And, you know, in some products like FPGAs or large DSPs or even large processors, you know, those termination resistors can be effectively incorporated into the design with active termination. But you do see that quite a bit that people will just put down the termination resistor in the wrong place. And that's something that's quite unfortunate as well, because it can't be changed. It can't be changed, you know, post layout. Right. You know, lesser problems, you know, can be like actually sometimes another very common one is you have a first order RC filter or maybe even a second order LC filter on a line. I know that I could either radiate noise out over a cable into the environment. So I should probably just bandwidth limit the energy that can come out of that cable. So you'll put down, you know, maybe a very small, you know, couple of peak fire capacitor to ground and maybe a one to 10 ohm resistor in series. So you get a nice little RC filter. But then you find sometimes I've had a couple of clients actually where they were a bit too aggressive with the value of the RC filter. And sometimes they confuse the termination resistor. They saw on a schematic somewhere, oh, I should have a termination resistor. Well, that should be 27 ohms or 33 ohms for some reason. I don't know why, but, you know, that's what we should put down. So they'll put down the 33 ohms and they'll put down a fairly large capacitor. And then, you know, they just don't get the bandwidth that they need, you know, the genuine bandwidth that they need for that signal. Oh, yeah, right. And I had it one time so bad.

Kieran Oleary: Some kind of thing is rolling off after some kind of quarter frequency.

Chris Gammell: Yeah, like I had a guy one time who said, oh, my audio product doesn't work. And it's kind of dovetailed nicely with my, you know, my various backgrounds bringing in the signal integrity and the audio stuff. And it works fine at 1K, you know, but at 10K, it's just doing nothing. And so I think I might know what's going on here. You know, I mean, the issue was presented as a frequency response issue. Like, oh, look, you know, I need it flat from 20 hertz to 20K. You know, I'm doing a really, you know, it's really important that I get the full audio bandwidth. And at 10K, I'm just, oh, it's terrible. And I say, oh, yeah, I don't even need to see the frequency response. I have a pretty good idea of what's going on here. But, you know, but they're much less problematic because if you put down, you know, a resistance to ground, if they put down a resistance to ground, excuse me, a capacitance to ground, and it's not populated, but it's there. Or they put a series zero ohm resistor, you know, that can be really useful because it helps, you know, if you have noise on the board, you have an RC filter, if you have some kind of radiated emissions issue, you know, like all these things. If you make things easier for yourself in the design phase, you know, at the schematic entry point, anything coming off the board, try to allow yourself options. And then, you know, you don't need to be, you know, people often describe, oh, all this stuff, it's black magic and everything. Like, I mean, and, oh, it's a dark art. And, you know, like quite often it isn't, it's just, I mean, the laws of physics don't change. And all that comes into play is, you know, the parameters of your design, you know, where they fit into, you know, into the universe, like, you know, really. And what you need to do to make sure, you know, to make sure that, you know, everything works, everything works nicely, like, you know. So, yeah, I mean, yeah, really the current flowing in loops and, and again, that's the same with splits, the signal crossing over the splits. That's, that is the same problem. Like, you know, the, the, the outbound signal crosses over the split, but where's the return signal going? You know, with a high speed signal, it tends to have, you know, it's just a natural phenomenon where it will, it will attempt to, the return path will attempt to follow as closely as possible the outbound path. And if you're returning, you know, if the outbound path crosses over split, of course, the return path can't do that. And that causes problems.

Kieran Oleary: And when you say splits, you were, you were talking about the power plane, you were talking about the ground plane as well?

Chris Gammell: Like, it could be either. Actually, it's one thing that maybe people don't think of so much too, that a power plane can serve as a return path just as, just as easily as a, as a ground plane. People often think, oh, my return path for high speed signal will have to be the ground plane. But, and for, you know, for a low speed signal, that may well be the case. But for a high speed signal, the return path, if there's, assuming that there's a plane there, will be the adjacent plane. And so if you take, if you take a, if you take a four layer stack up, typical four layer stack up in a PCB, it will be layer one will be a signal. Layer two will be ground. Layer three will be your power planes. And layer four will be signals again. Now, of course, there's variations and everything, but typically signal, ground, power, signal. And a high speed signal on layer four will have its return signal, a return path on layer three. Now, if layer three is, is just made up of patches of power planes, then that can be completely appropriate and completely fine and completely resolve absolutely fine with the laws of physics, provided perhaps that the power planes are sufficiently decoupled. But if you think about it, I mean, you know, in the past, people ask, oh, how can that be the case? And, you know, there's, you know, there's obviously, you know, good, good physical reasons why that's the case. But if you like to think about it, and it's perhaps not entirely accurate, but it does kind of get the point across that if you differentiate a constant, you get zero. So if you differentiate five volts and it's a constant, perfect five volts, then it's zero. And zero sounds a lot like zero volts, you know? So that's, you know, very, very brief summary. You know how you can perhaps describe it kind of orally, you know, in one sentence. And it's perfectly valid for the return signal to be on layer three. In fact, that's where you want it. Again, you want to keep the loop as small as possible. So if the signal is on layer four, you want to return return path to be on layer three for a high speed signal. So if you chop up layer three into lots of different islands of copper, because there are different power planes, five volts, 3.3 volts, 1.8 volts, then unfortunately, the return path can stay adjacent on layer three to the signal on layer four. And then it will attempt to find a way elsewhere on the board, you know, to complete the path. And then you start increasing the loop size and you increase the overall loop area. And that has a whole bunch of bad side effects. So, again, it comes back to the light bulb. You know, you've got a light bulb, you've got an outbound signal and you've got a return signal. Where is that signal going? And earlier when I said either you or James Clerk Maxwell are going to choose the return path, then you get down to, you know, the laws of physics and Maxwell's equations and everything, which will determine where that return path is. But you kind of you want to take it out of his hands and put it into your own hands as best you can, you know, within reason.

Kieran Oleary: Yeah, that's great. Well, I think that's probably a good place for us to wrap up here, too. I think I saw on your Twitter that did you sit on Twitter? You said you have a Maxwell statue in your hometown.

Chris Gammell: Oh, yeah. Oh, yeah. Something about Maxwell. I'm a big, big Maxwell fan. So James Clerk Maxwell was Scottish and he was born in Edinburgh. So the James Clerk Maxwell Foundation is here in Edinburgh in India Street. I think it's 14 India Street. And you can go and visit his house where Maxwell is born. So if you're ever in Edinburgh and you contact the James Clerk Maxwell Foundation, they have a museum set up to his life and works and they have some really great exhibits there. And, you know, and you just have to, you know, they're really passionate, enthusiastic guys, you know, and the you know, you see some really interesting things there. And, you know, there's a plaque on the wall, you know, and the IEEE have, you know, made a statue on George Street here in Edinburgh with Maxwell and his dog. And you see Maxwell's equations, maybe on my Twitter feed, if you look back, you'll see Maxwell's equations on the ground in George Street. And a couple of years, I mean, it's been there less than 10 years. It's been there since, you know, since I've moved to Edinburgh. But it was quite depressing at the start. I mean, he's quite often, you know, considered the greatest scientist between Newton and Einstein. But very few people in Scotland knew who he was. And that's partially down to, you know, the nature of his work is maybe quite abstract for a lot of people. A little hard to grasp. Yeah. But this summer, there happens to be an Italian restaurant just near the statue. And I was standing outside the restaurant waiting for, you know, waiting to meet somebody. And in the 10 minutes that I was standing there, I saw two fathers at different times, you know. So I saw two fathers with two daughters, actually. And both of them were coming along. They were tourists. And they were fathers explaining to their young daughters who this guy was. And the daughters were getting their pictures taken, you know, with this guy, Maxwell. Which was amazing. It was. Heck yeah. Yeah. It was great. I mean, it was even 10 years ago. I mean, like, I couldn't imagine. But now I was. And I thought my wife. Like, it was great. Like, it was, you know, they weren't sons. I mean, they were daughters. Like, you know, the women in tech thing is very important and growing. And it was just, it was quite pleasing to see, actually. So, yeah. Yeah. He's. Yeah. Yeah. He's. Yeah. So, I mean, the statue's there. You can get your. On George Street, just near St. Andrew's Square. And, yeah. Like I say, on India Street, you can go and visit basically a museum to him as well. And, yeah. You know, a lot of stuff happens here in Edinburgh. But actually, one last point on that is where Maxwell was born is just around the corner from where Alexander Graham Bell was born. And it's. Oh, really? Yeah. And it's just around the corner from where Peter Higgs lives. So, the Higgs boson. So, you know, like if you are, if you do happen to be in Edinburgh, just go down there and just start drinking the water and breathing the air. Like, you know, just do everything. So, I mean, I mean, a lot of this comes back to the Scottish Enlightenment. You know, like, I mean, you know, 150 years ago, 200 years ago. And, you know, the, Edinburgh was, you know, real center of learning, you know, from Western philosophy and economics and science. And, you know, I mean, Lord Kelvin, you know, was over in Glasgow, which is an hour away. And the Harriet Watt University in Edinburgh is named, excuse me, is named after James Watt. You know, he's also a Scottish guy. So, yeah, there's really a lot going on, you know. And actually, Dolly the Sheep was, you know, cloned or created or invented or whatever, you know, in Scotland. You know. So, yeah. You know, there's a lot of good stuff here. And there's an awful lot of mixed signal electronics as it happens as well. Like, you know. So, yeah, good little tech scene here. But, you know, it's only a city of half a million people. But a lot going on here. But, yeah, I'm quite a big Maxwell fan, which ties in nicely with my work and does go back to when I said, you know, at the beginning, kind of, you know, wrapping things up nicely is my fine year project where I felt like such a fool where the robot took on a mind of its own. I said, OK, I'm going to get that. I'm going to get on to the bottom of this. Like, you know.

Kieran Oleary: Yeah. Yeah. Yeah. So, so if people are stopping through Edinburgh, where can they, yeah.

Chris Gammell: How can they contact you and where can people find you online? Well, OK. So, on Twitter, you can find me at O'Larry Kieran. So, O-L-E-A-R-Y, Kieran, K-I-E-R-A-N. And, or the company account is Mixed Signal Sys. So, Mixed Signal Sys, also on Twitter. Or the website is www.mixedsignalsystems.com. And there's a, you know, contact us form there. So, yeah, absolutely. By all means, if you're in Edinburgh and you want to see the technical sites, you know, you get literary walking tours and whiskey walking tours. There's no reason why you can't have a technology walking tour. There's enough here to do it. Like, so it'd be great to catch up with somebody if they're in town. Yeah. Yeah. That's great. Can I just do one really quick shout out, please? I won't keep you. It's just two seconds. Sure. Sure. Yeah. So, I think we spoke, you know, in 2015. So, I was the general chair of a technical conference, EMC Compo 2015. And it's, the full name of it was, you know, the 10th International Workshop on the EMC of integrated circuits. So, generally, EMC is thought of as a system level thing. And absolutely it is. But nowadays, there's more and more of a focus on EMC at the IC level because there's more and more integration. You need to address some of these issues at an IC level, you know, like rather than at a system level, just because so much more of the system is on ship. And coming up later in the year, the 12th International Workshop on the EMC of integrated circuits is taking place in China. So, I can't remember the exact details, but it's in Shanghai or just outside Shanghai in September or October. And if you go to www.emccompo.org, I'm sure that the details will be there. And it moves around the world. It was in Edinburgh in 2015 because it was the 150th anniversary of the publication of what essentially was Maxwell's equations. But, you know, it moves around the world and it'll be in China. So, anybody in Asia who wants to pop along, I'm sure those guys would be very happy to see you. So, and, you know, you can learn a lot about EMC and everything.

Kieran Oleary: Looks like it's coming up in October. I just found it. It's October 21st to 23rd in Hangzhou, China.

Chris Gammell: Yeah, in Hangzhou, Hainan is just outside Shanghai. But, I mean, like, typically the guys that attend this would be, you know, aerospace companies, automotive companies, you know, like because, you know, typically the guys, aerospace companies, you know, like aircraft companies, you know, Boeing, Airbus, all these guys, you know, the automotive companies, the military companies, because they want to, you know, they want absolute EMC performance. And in the consumer space, you can get away with a lot and deal with it at a system level. But these guys, with more and more integration, addressing more and more of these issues at an IC level is very important. And I know how much work was involved in doing the conference in 2015. It was very rewarding, but it takes a lot of work and, you know, it would be nice to help these guys out in China. And if you can, you know, if you can solve your EMC problem at the IC level before it even touches the system or certainly approaches anywhere near your release date for a product, you know, you'll be doing yourself a good favor. So, you know, if you happen to be in town or have the resources to travel to China, then, you know, I'd really recommend you go there if you can.

Kieran Oleary: Awesome. Well, Kieran, thanks for being on the show and talking about all this stuff. I definitely look forward to talking more about this stuff and keeping my loop small so that Maxwell doesn't come and bite me.

Chris Gammell: Yeah, oh, absolutely. Yeah, well, yeah, thanks. And thanks for the amp. I mean, it's fantastic. It's really good. You know, you just learn so much. Like, you know, you learn so much in university or whatever, but like really you have to see what people are doing and what they're up to and what real life problems they encounter. And it's fantastic. Yeah, well, thanks, man. Yeah. All right. We'll talk soon. Okay. I'm off to bed now. Okay. Thanks, Chris.