#579 – ADC Chip Design with Anthony Wall

Anthony Wall: This is the Embower Podcast. Release February 27th, 2022. Episode 579. ADC chip design with Anthony Waller.

Chris Gammell: Welcome to the Amp Hour. I'm Chris Gammell of Contextual Electronics. And I'm Anthony Wall. I'm doing a PhD in mixed signal design, designing current mode circuits, current to digital converters for high precision applications. Hi, Anthony. How are you? I'm good. It's an honor to be on the show finally. I've been listening for years and I suppose maybe one of the first generations of people who were listening to the show maybe before undergrad. And it kind of formed my undergrad experience and, you know, learning about electronics in a more casual way like this. And yeah, it's an honor to finally be on the show on the other side.

Anthony Wall: Well, that's great. I appreciate that. I appreciate listening. Some of the words you said, obviously you have surpassed much of my knowledge already, given the words you said at the beginning of the show. What are you doing?

Chris Gammell: Yeah, apologies. Yeah, that's the pre-programmed.

Anthony Wall: No, no, no. This is good. This is why you're here. We want the cool stuff. This is awesome.

Chris Gammell: Yeah, you know, you ask, you ask, you know, a PhD student like, well, you know, how are things getting on? Like, you know, and you just get the kind of, oh, you know, I don't want to talk about it. So you give the complicated, quick answer. I apologize. I gave you that instinctually.

Anthony Wall: So no, no, no, that's good. I'll break it. It's the stop asking me questions answer. You know, it's like, oh, I work on very complicated things. And unless you want to be here for a while, but we do want to be here for a while. So yeah, let's hear the other one.

Chris Gammell: Right. Well, yeah, right. We have an hour. So, you know, everything around us is analog. And it's obvious at this point, we need to sometimes convert these analog signals, whatever they may be, to digital. And I focus primarily on taking currents and converting them to digital, hopefully as efficient a way as possible in terms of doing it power efficiently and area efficiently on chip. So, so yeah, effectively using, using ring oscillators to do so. And I'd be, yeah, we can, we can, we can chat about more about how I do that. Sure. And so where are you, where are you doing this PhD? Yes. Apologies. Yeah. I'm in, I'm in Ireland. So I'm not sure if your podcast player has a, you know, a half speed option. So if I'm talking too quickly, you know, feel free, feel free to use that. I'm, I'm, yeah, I'm in Ireland. I'm in Cork in the Tyndall National Institute. It's a, it's a kind of a, an Irish government run and funded institute, which focuses on everything between atoms and systems. So from like kind of research in kind of both silicon and exotic material devices, like the next generation of transistors all the way up to, you know, embedded systems devices and their applications.

Anthony Wall: Oh, cool. So it's like a, like a base science. It's like the, the Bell Labs of the Irish government.

Chris Gammell: Yeah. That's a compliment. Yeah. We'll take that. Yeah. But effectively, yes, that's the idea. And, you know, you want as much cross collaboration as possible between the different groups. And that, that's the kind of idea for all in the same house together. Hopefully we'll talk to each other and, and, and come up with some, some good ideas. And I'm, I suppose I'm situated right in the middle of that. I design circuits using devices that someone has already, you know, created, you know, well-established technologies. And hopefully someone will then further use the chips I make in, in kind of systems level stuff. So we're, we're sitting right in the middle.

Anthony Wall: Yeah, that's awesome. And so interesting. So it's like a national university undergrad all the way up through grad programs like you're in.

Chris Gammell: No, actually. So it's, it's strongly, so it's associated with all of the, so all of, all the universities in Ireland are public. Yep. And it's, it's, it's strong and it's strongly associated with university college Cork, which is right next door, but it's, it's, it's affiliated with all of the universities in Ireland. So I did my, right. So I did my undergrad in university college Cork. And then I was whisked straight, straight across the road into Tyndall to, to, to, to be told, go away for a few years and create some circuits. So I did my post-grad in, in Tyndall. I'm doing it.

Anthony Wall: That's great. No, that's really cool. And so we were talking a little bit about before the show, just about like, obviously we've been talking about Intel and all the fabs that are happening in the U S right now, driven by a lot of like tax policy and just wanting to reshore some of the semiconductor stuff. But Ireland has a ton of, of semiconductor stuff. Notably analog devices is really big there. Intel has a big fab there. Why, why Ireland?

Chris Gammell: Yeah, it's, it's an interesting question. I'm not, I'm not sure of the number, like the specific numbers on it, but it certainly feels like the highest, you know, proportion of, of semi companies to people in the world, you know, area wise, outside the Silicon Valley, of course. It's an interesting story. I could kind of trace it back to the, the 1930s. So, so when, when aviation started becoming really, really popular, the first transatlantic flights started happening, but flights couldn't go from like London or Paris to New York on a single tank of fuel. So they'd have to make some refueling stops. Yeah.

Anthony Wall: I always think of the, the, the map view in Indiana Jones where they show all the difference. It's like, why are you doing all these little hops to islands or whatever? It's like, oh yeah, those planes, they were not fun. Right. A lot of shots drinking on those, on those flights, I think. Yeah.

Chris Gammell: And thankfully we were one of those stops, right? You know, there's a convenient little Island between, between London and, and New York. I think, I can't remember exactly what the range on these, these flights were, but it was just slightly too far away for London. But Ireland was, was right in the sweet spot. So. Interesting.

Anthony Wall: Yeah. I always think like they, they go to like Newfoundland, like if you're going from New York, you'd like go up to Newfoundland and then maybe over to Iceland, like Reykjavik. And then maybe down to Ireland and, and then over to the UK. Exactly. Or over to England.

Chris Gammell: Yeah, exactly. Yeah. They were, they, they were actually the hops that were used. So the, the Irish government with the British government and the US, they all came to this agreement that they'd build this airport in the middle of nowhere in the West of Ireland. And, you know, if, if anyone's familiar with Ireland, that's a really kind of barren place. Like the population density is so low. Was it just like farmers and. Yeah. And the cliffs or what? Actually. Yeah. Farmers and cliffs. That's, you know, really beautiful part of the world. Yeah. Totally. Incredibly beautiful, but yeah, very, very, very, very sparse. Yes. So it was actually, you know, the, this town of Shannon built up around this airport. And it was the first town built in Ireland, like the first new town in Ireland in like 300 years. So it was like, you know, you know, this, this big thing. And this was kind of the 1930s. So people would stop in, in Shannon and, you know, they'd be there for a couple hours while the plane is being refueled and whatever else. And off they'd go again. So the only experience people would have with Ireland would be the meal that they'd have while they're here. So, you know, this, this fella, this hotelier by the name of Oregon, you know, he got the idea that, you know, at the time, I think it was British Airways were supplying the food and, and stuff at Shannon. And Ireland, having gained independence only 10 years previous, obviously didn't want, you know, the, the only experience people have with Ireland be, be, be a British one. So they said, right, this Oregon fella says, I'll, I'll, I'll start my own little restaurant in, in Shannon airport and people can have Irish food. But this guy had, I think he trained in Paris. He had an eye for detail. And one day someone was bringing out a cup of coffee for, for someone. And, you know, he says, this looks very plain. So your man goes away and puts cream on top and whiskey in it. And that was the story of the Irish coffee, right?

Anthony Wall: Our listeners may be familiar with the finer point.

Chris Gammell: Yeah, actually. Yeah. Yeah. No, I had actually never had an Irish coffee until my American girlfriend came over and says, oh, you know, you've never had an Irish coffee and you're Irish. Like anyway. So, yeah. So, so this guy really people raving about this restaurant, long story short. And, you know, the Irish government gave this Oregon guy a big pat on the back. So he kind of gets this mad idea, right? How can we, how can we hold people in Ireland a little bit longer? And he, he invents the idea of like a duty free shop where people don't have to pay tax. And, you know, they sold lots of Irish goods and that was kind of where duty free came from. So then he said, right, let's build a fence around the airport. And instead of having one duty free shop, maybe we can have like many duty free businesses. And, and his idea, he kind of put it to the Irish government, right? Anyone inside this area, as long as they can fly in the raw materials and fly out their processed product. As long as they can do that, they don't have to pay any tax to the Irish government. The idea was that they, you know, create jobs and stuff like that. So, so you can imagine, right, that, that benefits certain industries. Right.

Anthony Wall: Well, this is starting to already sound similar to some of the tax incentives driving just chip design generally these days. So, so, yes. So fast forward a little bit here. So, so basically I'm guessing the yada, yada, yada is chip companies did that.

Chris Gammell: Yep, exactly. Yeah. So, so ADI came along and they built a fab here for that exact reason. It's easy to fly in wafers and easy to fly out chips. Right. So, and, and it worked really well. It created some, some high quality, well-paid jobs. And at the time, Ireland was a very poor country. So that was, that was a big boon for the economy in the area. You know, happy days. Yeah. And, and, you know, once one comes, the rest follow. We're also really, really lucky. So the story of Intel and LeakSnip is, is, you know, similar. It's like, I mean, ADI are here. There are people kind of coming along. There are more graduates, da, da, da. And also the, the other big benefit for us is a rock. We, we come, we, you know, we're in a limestone region. So like, you know, the rock is really porous and it kind of absorbs vibration. So there's very little seismic activity here. So if you're building a fab, right. If you're trying to build a fab and you have, you have these little shakes and you're trying to align your lithography machinery, it's not, not going to go so well. I assume not, not my area of expertise.

Anthony Wall: And that's always interesting. Cause I've heard that, you know, obviously they talked about that when I was at Samsung and I get that, but then I'm like, but what, what the hell is Japan doing? For years. I mean, not as many like, like newer fabs, but like, but still like a lot. And I think Taiwan has some activity too. So they must just have them on shaker tables or something like that, or anti shaker tables.

Chris Gammell: I assume so. Or, you know, maybe it, I, I don't know if it's a yield thing or if it's a, if the ground is shaking, stop making chips. I'm not, I'm not too sure. Yeah. Yeah. Yeah. I assume it could be a yield thing. The ground shakes. You're like, I guess we're throwing this batch out as long as you detect it fast enough. Right. Yeah, exactly. Yeah. So that, that, that they're the two main things that kind of attracted the first two were kind of Intel and, and Qualcomm, sorry, Intel and, and ADI. And, you know, now we have, you know, the who's who of, of semiconductors. We've got, you know, on semi and Xilinx and Qualcomm and uBlocks and whoever else, you know, it, once one comes, they all come.

Anthony Wall: Yeah. And well, that, and that's the effect that we were kind of talking about too, of like, you have this talent pool now, and then it's like, all right, well, now we want to hire people. And it's such a specialized talent that, you know, a Qualcomm comes in and says, oh, well, we could hire a bunch of, we could, you know, quote unquote, set up an office in Limerick and hire, you know, four designers. Maybe it's only four designers to start, but then that's like a toehold. And then it becomes 10, 50, a hundred, whatever. And then it just becomes this bigger and bigger thing. And there's such knowledge around it.

Chris Gammell: So, yeah. And yeah, it's, as you said, you know, the, the tax incentives, the Irish government at some point, obviously you went, oh, you know, these, these semiconductors, they might like catch on or something, you know, let's, let's, let's give some tax incentives. And also I think the universities were quite smart in terms of like, you know, all the analog stuff seems to be down in Cork. So let's teach all the Cork students analog stuff.

Anthony Wall: Well, that's, that's what I was wondering then. So then Shannon is pretty far north of, and Limerick are pretty far north of Cork. So how do those get connected then?

Chris Gammell: Well, I suppose. Maybe my map skills are off. I don't know actually. No, you're right. You're right. It's kind of, there, there's a kind of a sense of like, you know, Dublin and not Dublin. So I always think of people in Dublin, they do digital stuff. I don't know what they're at at all. And down in the country, we're kind of a bit rougher. Like we like the analog stuff. It's a bit messier, you know?

Anthony Wall: Yeah. Yeah. Yeah. So it's the real world.

Chris Gammell: Yeah, exactly. Yeah. Yeah. You know, do this, we're going out for a hike. Exactly. I'll be killed for saying that, but anyway.

Anthony Wall: Yeah. But, but even the fact that Dublin then does have, you know, so just now the whole country, you know, you've stated now three different regions of the country, pretty much, you know, West, East and South, I'm not sure about the North, but like, okay, so now you have three different areas of doing chip design. That's pretty significant. You know, that's, that's, that's not, that's not small for, for a pretty, pretty small country. So.

Chris Gammell: Yeah, absolutely. And, you know, there are very few places, you know, I'm faced with the fact, you know, I've done my undergrad in Cork. I've done my postgrad, hopefully now soon in Cork. You know, I would like to see the world. And there, there's the urge to go to Silicon Valley, but it's, it's a hard sell to get me to go anywhere else because you can have a job in Ireland, you know, it's, you know, there's a list of companies that are, that are looking for people here. So yeah, we're, we're very lucky.

Anthony Wall: And you can't beat the weather either, you know. Sorry, I was actually, I muted myself and I made that joke and I was like, oh, that didn't land.

Chris Gammell: No, I, there was, yeah, that, that is.

Anthony Wall: I mean, Irish weather is like so similar to like Cleveland and Chicago that I feel like I'm allowed to make that joke in the old days, you know.

Chris Gammell: No, you, you are. Yeah. I, I've only ever been to Chicago once. It's, it's feckin' windy. Yeah. It's very windy. Yeah. Yeah. Yeah. I suppose. Yeah. They call it the windy city, right? Yes. That's right.

Anthony Wall: Yeah. Well, so, so you might travel after PhD is done, stuff like that. I mean, what is, what is the recruiting like out of, out of like Tyndall or, or more broadly just in, in Irish upper universities?

Chris Gammell: It's yeah. I mean, I think we have a similar problem here that, that, that you guys have, like it's hard to get engineers. Companies can't fill positions and out of Tyndall, you know, a lot of our, you know, a lot of our projects are part funded privately. So you'll have your sponsoring company may offer you a position and, you know, got it. If they don't, you, you walk outside the door and turn left and there's the other fella, you know? So, you know, it, it, unfortunately microelectronics isn't cool as cool as it was. I mean, I think it's really cool, but it's not, it's not AI. So it's harder to get people in the door.

Anthony Wall: Oh, interesting.

Chris Gammell: Yeah.

Anthony Wall: Yeah. I just feel like it'd be so specialized these days too. Like, I mean, there's so many programs. I mean, obviously there's a lot of AI type of roles that are popping up and stuff like that, but there's gotta be just such specialized knowledge. I, I have to imagine that eventually these chip companies are gonna be like, oh, I guess we had a upper salary, you know, just like the usual supply and demand type of stuff.

Chris Gammell: It is. I think that what I've seen in the last couple of years is the chip design companies have kind of said like, oh, you know, all of our graduate students want to do AI. So, you know, for example, Qualcomm and Cork have brought about this, this mixed signal AI group. And you know, that that's kind of exciting for, for young engineers. So there's kind of, you know, you can pay people more or you can let them do the jobs they want to do. And I think there's, you know, there's a confluence somewhere in the middle. Yeah. Yeah. That makes sense. I mean, chips for us, please. Yeah. Just please give a, yeah. Oh yeah. We'll design you chips. Like you're not going to get chips in the next two years, but there's people designing them. Right.

Anthony Wall: Right. Exactly.

Chris Gammell: Exactly.

Anthony Wall: Okay. So you're at Tyndall, you're making these chips. What, what drives then your research? So you, you're, you're working on these, you know, so you, they're called starved ring oscillators, starved current ring oscillators, right? Right. Current starved ring oscillators. Yeah, sure. Current starved ring oscillators. Yeah. So that's someone who needed that specifically. And then they kind of like put out a, a call for proposals for that sort of thing. Or is it like, you're just because of the, the, the PI that you're working with is, is already has this research and needs to take it forward.

Chris Gammell: Like what, what drove you towards that, that space? Yeah, that, that's a really good question. And you've kind of touched on our dirty little secret in the, uh, in the ADC research world, I suppose. It's great to have an application, like a killer application that your chip is the only chip that will work for this particular thing. Unfortunately, that that's tougher and tougher to do it. You know, ADC research is so mature at this point that most things have been done in terms of, you know, so a lot of it comes down to like, oh, you know, I can, I can do this conversion for X number of picodoules or, you know, I can do this conversion. Faster. Yeah.

Anthony Wall: Yeah.

Chris Gammell: So it becomes a race to the bottom in terms of just, you know, figures of merit and performance. So I was lucky enough that there was a company that, that, that wanted to hire me. And, uh, I said, for some reason I wanted to, to research more. So they said, look, you know, go away for four years, do something fun. We'll pay for you. And, and, and if, if the research is valuable, great. As I was growing up, I was always really curious about, you know, how capacitors worked. I never understood how capacitors worked. So, uh, I now, I now, uh, have ring oscillators with capacitors in them and I love playing with them all day. And that's effectively why I'm doing what I do. Nice.

Anthony Wall: That's great. So do we need to then maybe talk about how ADCs work? Would that be helpful for this conversation or we kind of just kind of hand wave past it?

Chris Gammell: Yeah, no, I think it's a, it might be interesting to kind of just understand why we need, you know, well, right. Why do we need ADCs? Maybe that's clear. Why do we need current ADCs and why do we, why do we need to keep researching them? That's maybe more interesting question, right? Yeah.

Anthony Wall: And what, what is the actual application of, sorry, what is the usual target? Application for current ADCs?

Chris Gammell: Uh, it's, yeah. So, so there are more and more applications for them recently. Uh, for example, like a lot of electrochemical sensors. So, right. So you have, you have some sensor and whether it's, whether it's pressure, whether it's, whether it's giving you an out. So you have a sensor sensing something and whether it gives you a pressure or a, or a flow proportional to what you're measuring dictates, whether it's a current or a voltage ADC that you need. Right. So, you know, if your signal is, for example, a microphone, uh, that might get, that might be a capacitance. So you might get a voltage out. Whereas if you have an electrochemical sensor, you might have some kind of ionic current you're trying to sense. That is a current. And, and you want to use the type of ADC that your signal is proportional to. Like a photo diode gives you a current proportional to your, your, your light. So, so if you were to use a voltage ADC, you'd, you know, you'd either have to do some conversions or deal with horrendous non-linearity, you know. Right.

Anthony Wall: Or have a really expensive resistor. That is a, a current to voltage converter, right?

Chris Gammell: Right. Yeah. That, yeah. I, I make really expensive resistors.

Anthony Wall: Yeah. So obviously Keithley, that was my, my jam at Keithley a little bit when I was doing stuff that was, you know, a lot of what the measurement equipment is, you know, you're measuring current and it's just going into a feedback circuit as going through a fancy resistor, right? It's that same kind of, you're doing a current balance with a fancy resistor in between the two. And so it's like, okay, well, that resistor is, is, is part of that conversion matrix,

Chris Gammell: I think. Exactly. Right. So, so, okay. So why don't we take that as the basis? Like, okay, you can now measure currents, right? You put it into a TIA, a transimpedance amplifier, you know, that sounds fancy, but it's effectively an op amp with a resistor in its feedback path. Right. And the idea is you, you feed your current into that and the current develops a voltage across the resistor and that's your, that's your output voltage. Yeah.

Anthony Wall: And actually the photo, so people think about like a photo diode conversion circuit, that's often what that is, right? I mean, it's, you have a photo diode on the left side of the op amp, you know, left or right saying here. And then on the right side, you got the output of the op amp. And in between you get that fancy one mega ohm resistor. Exactly.

Chris Gammell: Right. So if, and you know, you have a voltage out then and you can put that into your, uh, your ADC chip, right? Your typically ADC chips will measure a voltage and give you a digital code out and you know, happy days. That's, that's job done. So the, the question then is right. Okay. I now have an op amp on my board with a resistor in the feedback path and I have an ADC chip. What if I want to put that all on one, on one monolithic chip, you know, one, one chip integrated all. It becomes a lot more difficult because firstly, you know, you might, you might get a nice op amp based on some old BJT technology that will give you a plus minus 15 volt rails.

Anthony Wall: Yeah.

Chris Gammell: Right. That'd be lovely, but. Make sure that's a nice linear too, you know, really low noise.

Anthony Wall: And right.

Chris Gammell: Yeah. Yeah. Like no matter how much money, you know, you can pay loads of money for that chip and get it, but like, no matter how much money you have, you can't put that on your own chip. For example, right. I, I designed a chip on 65 nanometer and that will give you a 1.2 volt supply rail in, in the core of the chip. So yeah. Yeah. You know, pretty much the same as plus minus 15. Yeah. Yeah. Like do whatever you want as long as it's, you know, between zero and 1.2 volts, which is that's, that's, that's only as so many VGSs of transistors. That's a, that's a, you know, right. So that's, that's problem number one. That that's why it's tough to build a TIA. A transimpedance amplifier on a chip. The second thing is right. You have this resistor in a feedback loop. Like how do you make a resistor on a chip? Well, you must have this resistive material and you need to snake it along and make a long path of resistive material. You know, if you're trying to measure a small current, you need a large resistor to develop a large voltage across it. So then your resistor just becomes huge. It's this big, long snake and you're paying for a chip and most of your chip is resistor. Like, I mean, it, you know, it, you then literally have the most expensive resistor in the world. Yeah.

Anthony Wall: Right. I just imagine you, your, your graduate thesis being like, well, it works, but we only get one chip per wafer. A hundred percent yield.

Chris Gammell: Yeah. I drew the snake in Microsoft paint and there we go. Exported the GDS files off to the fab and happy days. Perfect.

Anthony Wall: That sounds about, I might be able to do that. Like if I went and took like Matt's course, Matt Ven's course and like how to use GDS and stuff like that. And maybe I could maybe get to that point, but that's about it.

Chris Gammell: Absolutely. Yeah. You know, you could, you could make quite a neat, say you have, you know, say, you know how to make a differential pair and you might, you know, you know how to make an output stage of an op amp. Right. So say you can make a traditional op amp. You can make a TIA. You make a big resistor. You can make a TIA. Yeah. But then that's the question of why are we still researching these things? If we know how to make them, you know, we want to make them small.

Anthony Wall: Well, and I think some, so it sounds like some of this is logistical cost, all of that kind of stuff. But then I think about the, you know, like I think about a lot of the analog circuits I've worked on in the past too, and they're always plugged into the wall and they're, you know, and they're, they're already at a, you know, 30 C ambient, whatever the warmup time is. And it's just like, okay, well that doesn't work great when you're trying to measure something in the field or your battery powered or whatever. So there's then that whole power consideration thing that I'm sure is part of your work as well.

Chris Gammell: Exactly. Like, you know, the guy who sits beside me, he, he's kind of more on the systems level and he goes out to fields every day and he, he, he checks for diseases in potatoes and cows using electrochemical sensors. You know, there aren't too many outlets in the fields of Ireland. So you need, you need the battery to last, right?

Anthony Wall: You either got to do that or build your own methane to electricity converter. Actually, yeah, right. You know, you got to strap it on the cow face and, and then, then you have a portable generator. Exactly. Yeah. Ubiquitous free energy right there. Yeah. Cause remember folks, it's not cow farts that are the problem. It's cow burps. That, that is the source of greenhouse emissions.

Chris Gammell: Exactly. So, yeah, I think we've just solved the climate crisis. Yeah. Yeah. So, so, okay. So it's clear look, right. That these, you know, it's hard to make a TIA on a chip. So, so what are your options? Right. So you say, right, well, my voltage, my available voltage swing has shrunk. Right. So the available voltage across that resistor has shrunk. So you, you can kind of make a hand wavy argument to say like, if you can't have as much voltage, you can't have as much signal, which affects your signal to noise ratio. So it's harder to make good quality analog circuits on a chip. And the ones you do make are expensive because you have these big long resistors.

Anthony Wall: So does this mean you also have, I mean, so you mentioned less, less voltage overhead as well. I mean, so I think about a lot of the op amps I've used in the past and they have, you know, like a protection diodes that are like just the standard 0.7 volt drop, you know, up to the rail. And so do you actually have those or, or no, do you have to go protection list?

Chris Gammell: No, no. I, I, as in you can, you can use, so the, the, the chip will have kind of a core region and it's, it's IO region. You can use the whole rail in the core. So you can use zero to 1.2 volts. You have protection diodes, but you know, they will allow you to go to, to, to the rail. Right. Okay. So yes. So yeah, you know, you can use the 1.2 volts, but the question is kind of right. What direction are chips going? You know, the, the voltages are shrinking, but what's maybe getting better. And you say, right, well, transistors are getting faster. And what does faster mean? Well, it means the, you know, the parasitics are getting lower. So the capacitances of each transistor are getting, getting lower. So that kind of begs the natural argument that, okay, rather than representing our signals in the voltage domain, i.e. turning a current into a voltage and then digitizing it, could we maybe turn a current into this kind of time domain thing, a time domain representation, and then digitize that? So, you know, you imagine, right, you feed a current into a TIA and it will give you a voltage out. Well, why not feed a current into a ring oscillator? And that will give you a frequency out, right? So, so if chips are getting better at generating frequencies with more precision, because the capacitances and the parasitics are getting lower, maybe we should go that direction. And that's the core idea behind, behind using say like ring oscillators for, for current measurement.

Anthony Wall: Okay. And so then can you give us a quick, quick, quick, quick refresher? I'm going to just say it. Can you give me a quick refresher on how a ring oscillator works?

Chris Gammell: Right. It's, yeah. I, I mean, you, you say give you a quick refresher. Honestly, sometimes I need a quick refresher. They're very unintuitive things.

Anthony Wall: Yeah. I remember there's like a feedback path. That's like, wait, why would you hook that there? I remember that with oscillators, but that's about it.

Chris Gammell: Yeah. And I mean, depending who you'll talk to, different people will have different explanations and definitions. Like you'll talk to an RF guy and, you know, they'll start telling you about Barkhausen criterions and 180 degree phase inversions. You talk to a digital person and they go, that shouldn't go there. And you know, I'm somewhere in the middle between those two, I suppose. Got it. Got it. Yeah. So, so the way I like to think about it is if you, if you simplify it right down and you imagine say like a, a current source charging a capacitor, and that's analogous to, uh, you know, like a tap feeding water into a bucket. Right. And how do you know how much water is flowing into that bucket? Well, one way of doing it is measuring how many times that bucket fills up in a given amount of time. Right. So if you look at it, if you look at the bucket for a minute and you've, and you're filling out water and you have to empty it three times, well then, you know, you have about, you know, three buckets worth of water per minute is coming out of that tap. Yep. So effectively your ring oscillator kind of automates that idea. So if you think now you, rather than you having to pick up the bucket and throw out the water, what if I were to put a, a ball cock on the top? So it, it basically detects when the water level reaches a certain level, right? So I'm pouring the water in and when the water level reaches, this ball cock triggers and it dumps the water out of the bottom of the bucket. Right. So you've got, you've got a ball cock and you've got a valve that, that kind of releases all of the water. So now you can kind of, you can measure how many times that, that ball cock triggers per minute. And that's the same thing, right? It's just automated. You can do that many, many, many times. The smaller your bucket, the more accurate you are in a given time period. So that's, that's the fundamental idea, right? That's what you call a relaxation oscillator in, in, in the technical terminology, right? You replace the water tap with a current source that you're trying to measure and you replace the bucket with a capacitor. So you're filling the capacitor with charge and you dump that charge when it fills up and you measure how many charge dumps happen per given time period. Right. Yep. So, so that's a relaxation oscillator. And then if you want to think about what a ring oscillator is, like if you look at the schematic for a ring oscillator, it looks like three inverters all connected in a loop together.

Anthony Wall: Yeah. That's, that's the one I've, yeah, that's like a Jim Williams schematics always had those in there.

Chris Gammell: Right. Exactly. Right. So how do I equate that to what I've just described in terms of the, the, the water bucket? Well, if you think the VDD of those inverters, right, the power supply of those inverters, if you think of connecting a current source to all of those in parallel, right, then you've got three inverters that kind of have these, this capacitance inherently at their, at their output. You can see that the, the current will be steered into the capacitor, the capacitance at the, at the inverters. So effectively what that looks like is these three stages of the, of the, of the ring oscillator in series. So, or sorry, of the relaxation oscillator in series. So you've got, you know, one bucket filling and then it'll pass to the next stage and the next bucket will fill. And then when that bucket fills, it'll pass to the third element in the ring oscillator and that bucket will fill. And once that bucket fills, it goes right back to the start. And you've got this effectively, you know, instead of the bucket resetting itself, the bucket actually resets the next one in the, in the chain and the next one and the next one. If that makes sense, it's a, it's a weird concept to describe without showing it.

Anthony Wall: Yeah. I mean, yeah, no, it definitely is a weird concept. I mean, like how much, how much control then do you have? So like, is it just like a characteristic frequency then based on that or, or something else?

Chris Gammell: Yeah, that, that's a good question. And that, that's kind of what makes these things kind of tough. So you've got three inverters and you imagine, right. What's the analogy of this ball cock, right? What's detecting the, the water level or the, the charge level? Well, that's the voltage, right? That, so that's the voltage that the cap charges to. Well, what detects that is actually the threshold of the next inverter. So once you reach the threshold of the next inverter in the chain, it, it starts a charge dump. Right. Cause the logic levels you're saying, right. That's the logic level threshold. Yes. Now the problem is that inverter, I mentioned that the VDD of that inverter is connected to the tap, right. It's connected to the current source. So, you know, the voltage on that node is going to change. So the VDD of the inverter is going to change, meaning the threshold of that inverter is going to change, right? So the very act of putting current into this system changes the, the level that the ball cock triggers on. So you've, you've got this non-linearity inherent in the system. Also, right. I'm telling you that you measure how many times the water dumps per, per second, for example, that's assuming that all the time is spent filling the bucket. But there's also some time spent emptying the bucket. And if the time spent emptying the bucket is significant, you know, your frequency is kind of, you know, it's not only dependent on the filling, it's also dependent on the emptying. So your frequency slows down a little bit and you've got the, this kind of complex interplay between the amount of current you have flowing in and the VDDs and the empty time. So yeah, it's.

Anthony Wall: But that's interesting too, because then that starts to sound like a sawtooth wave, which is like the basis of a dual slope oscillator or sorry, dual slope ADC as well. And so that starts to make my mind think like, oh, okay, I could see how this might start to play towards an ADC. Yes. Application.

Chris Gammell: Yeah. And exactly. Like, I mean, a slope ADC is a very, very similar concept. You can, you can make the relaxation oscillator analogy and the water bucket analogy about a slope ADC also in a slightly different way, but you, you can make the same analogy. So, so you can imagine, right. Ideally you want a sawtooth out of this and, you know, effectively what I'm trying to do in my research is make it look as much like, like a sawtooth as possible.

Anthony Wall: Oh, interesting. Yeah.

Chris Gammell: So if you simulate one of these things, you'll see it roughly looks sawtooth ish, but the final value of the sawtooth is poorly defined and the, you know, it's not a, it's not a bang straight down after the end of the ramp. It's kind of a gentle slope down. So you want, I I'm trying to make the, the, the final point of the sawtooth well-defined and I'm trying to make the bang straight down so that there's no time wasted in the, in the, in the gentle, in the, you know, in the discharge. Yeah.

Anthony Wall: And I can imagine that you could maybe, you know, try and cheat a little bit and like have like a really low impedance path out of there to try and dump current out. But I would imagine that the frequency that you're ultimately like getting in or out of there would also then end up impacting how fast it wants to empty anyways.

Chris Gammell: Exactly.

Anthony Wall: Yeah. Yeah. You, you, you must be chasing your tail quite often on this stuff, huh?

Chris Gammell: Yeah. Yeah. It's, it's a fun one, right? Because, you know, with an op amp, for example, the tools are, you know, I, you know, you look at for LTSpice, right? It's really easy to do like a .ac sim on an LTSpice and look at your, look at your amplifier and say you're doing an op amp, a traditional TIA, say it's really easy to do an AI. Oh yeah. This is, this is how this thing works. There's a frequency response. You know, I can do a DC sweep and I can see where, where the TIA works and then I can do a transient, da, da, da. With this, it's kind of like, what does it mean to look at the frequency response of a ring oscillator? It's a, you know, it's this kind of, it's not a stationary circuit. It's always kind of rolling over. So just even from the tools point of view, it's kind of hard to see where these nonlinearities are coming from. It's hard to see where these non-idealities stem from, if that makes sense. Yeah.

Anthony Wall: Yeah. Yeah. Yeah. I could see that. I mean, how, how do you then start to chase those down? I mean, so I, I imagine that the impact is still the same. You still are like, oh, well, this needs to, needs to have some kind of predictability. So then is it like, do you have things that you can go after and. Right. What tools do you have at your disposal then? Like, you know, so you're designing this actually on Silicon, so you might have some more capabilities in there, but is it just like patching, patching, patching with, with different tricks or process differences? I don't really, I don't know what tools you have available to you.

Chris Gammell: Right. So, so I've kind of, I've kind of modified the ring oscillator slightly. So I've done some architectural stuff and then obviously there's the, the whole, you know, all of the tweaking that needs to happen. On the architecture level, I've made basically a good valve, right? So I can ensure that the dump happens really quickly and the dump is independent of how much current is coming in. So I know, right. So I know that my dump right now takes 70 picoseconds. I can guarantee you that. So if, if I know that I can also quantify the non-linearity that that generates, if that makes sense. So my frequency is a product of the charge time, but also this 70 picoseconds that we don't want, but I know it's there. So I can subtract it maybe digitally afterwards, for example.

Anthony Wall: And do you, when you say 70, 70 picoseconds, that is a limit or that is a precision? Like that's a guarantee that it's always 70 picoseconds.

Chris Gammell: That is 70 picoseconds on a, on a nominal chip. Right. So. Okay. All right. In, in the research world, like, you know, when you make things and you have to like make money and deal with like shareholders and all that crack, your chips have to work. Yeah. Yeah. Yeah. Idiots. For us, you know, if our, if our chip works on a sunny day in the lab, fantastic. You know, it happened. It has to work. This is going on a poster. Yeah, exactly. Just, just like nobody breathe. Let me just like write my paper here and now, get these results and put them into the conference. I gotta get out of here. You know, just turn off all the lights in the lab. Make sure nothing is.

Anthony Wall: I'm so sick of eating ramen. Yeah.

Chris Gammell: Yeah. So, so long and short, right? That 70 picoseconds that I'm mentioning, that is the time. So, so I have this, I have this switch that dumps out the charge, right? In reality, that's a transistor. So I know that if I give the gate of the transistor that a good blast, I know it will take 70 picoseconds roughly for that charge to dissipate. Now, will it take 70 picoseconds at, you know, minus 100 degrees Celsius, or sorry, minus 30 degrees Celsius or plus 100 degrees Celsius? No. Right. So, so that, that's some kind of future work I'll have to work on to try and, you know.

Anthony Wall: Yeah. You characterize it. You'd, yeah. I'd count for that sort of thing. Right.

Chris Gammell: Yeah, exactly. But I know it's roughly 70 picoseconds, right? So, so if I can subtract off that, I'm doing well. Okay. So, so, so yeah. So, so our first test chip was to kind of try and prove this model to prove that if I have this linear charging slope, and I know if I've constrained the maximum, kind of the, the, the maximum threshold, I suppose, and I've constrained the, the, the chart, the discharge time, if I've constrained those two things, does my model fit? I know what the non-idealities should be. And by the way, I'm talking about non-linearities and non-idealities. What this means is the, the, you know, the more the threshold shifts and the more the discharge, the longer the discharge time, that means my transfer function is non-linear, meaning the amount of current is no longer exactly proportional to the frequency. It's slightly less. The frequency is slightly kind of lagging the, the input current. And why I care about that is if you feed a sine wave into this thing, you'll see harmonics pop up because you don't have a, you know, Y equals MX plus C slope anymore. You have some, you know, X squared terms and X cubed terms. So, so when you do the maths, you'll see, you know, you'll see harmonics at, at multiples of the, the input frequency. So we don't want that.

Anthony Wall: Got it. And, and just to level set a little here. So the end application is just in broad buckets, is slow or fast?

Chris Gammell: So this will measure kind of, you know, nanoamp level currents at, at kind of megahertz bandwidth.

Anthony Wall: Okay. So pretty fast. Yeah. So not like, not like a Delta, 24 bit Delta Sigma of slow, you know, averaging over 45 days or whatever. Some of the, some of the ADCs want to do these days.

Chris Gammell: No, it's, it's, yeah, no, it's, it's definitely not that. It's kind of, it's mid precision. You know, it's, it's kind of eight bit precision right now. So we hope that, so, so we've proven the model with this chip and it's getting eight bit precision, but once we add the correction for it, we, you know, we hope we could add two bits and remove a lot of the power consumption from it. Yeah. Yeah. Yeah. So what, what's it useful for? I don't know. And if any listeners need a, a one megahertz nanoamp level.

Anthony Wall: This is actually the reason Anthony is on the show right now. He's actually shopping this around.

Chris Gammell: He's yeah. Yeah. Exactly.

Anthony Wall: Like, it's like going to Broadway and being like, Hey, you know, I got the script. I just need some dancers and a stage.

Chris Gammell: Yeah. Yeah. No, I'm, I'm, I'm one step away. Like every Thursday we go to a farmer's market to, to get lunch. And like, I'm one step away from setting up a stall and selling that chip there. I need, I need someone to use this thing, you know? Yeah.

Anthony Wall: Yeah.

Chris Gammell: Get your fresh chip designs here. Yeah. And I mean, look, you know, I've got a hundred of these things. I don't know how many people in the world can say they have a hundred chips right now.

Anthony Wall: So, uh, yeah, exactly. Yeah. Get on, get on to me.

Chris Gammell: Yeah. And actually, you know that. So, so we did our test chip. We taped it out last April and we were hit fairly severely with this whole thing as well. Yeah. I can actually talk you through how the, how the whole, you know, how we get chips in our world. Yeah. That'd be great. That'd be great. Or don't get them in, in, in this case. So the way it works is effectively. I, I designed my chip in, in, for example, in the cadence. I, we used the cadence tool. So I, I effectively end up with a set of GDS files, which is not too dissimilar, uh, broadly speaking to Gerber files in PCB world.

Anthony Wall: Yeah. Yep.

Chris Gammell: And, and I, I need to send those away. And, and effectively the way it works is if I was to buy it, if our group was, were to buy a whole mask set for an ADC, uh, for, for, for, you know, for a whole wafer, you know, it would cost at 65 nanometer, a couple of hundred grand. So that's not going to happen. So you have these companies that offer what are called NPWs or multi-project wafers. So a bunch of research groups will get together a couple of times per year and say, you know, we commit to buy, uh, uh, four by four millimeters on your silicon. You know, we, we buy this space on the wafer and we'll get a hundred chips back and we'll share that wafer with every other research group around Europe. So this company in Europe is called Europractice. I think there's a similar thing in the U S called Moses. Yep. That's right. So, so they, they then farm that out to the fabs and say, oh, look, you know, we have, we'll, we'll, we'll, we'll put a mask set together for this. We'll share the, however many hundred grand of tooling costs for that particular mask set. And we'll send the design off to TSMC and they'll give us back, you know, X number of wafers. So, so, so that's what we did, right? So the standard NPW is four millimeters by four millimeters. And we're, you know, we don't even, we don't even need that. I, I only needed a millimeter by a millimeter. So a couple of times per year, they'll offer a special, uh, a special setup where they subdivide the NPW again and they offer you a square millimeter. So, so that's what we got. And it's actually, you know, it's, it's almost affordable. It's, uh, three and a half grand.

Anthony Wall: That gets you more chips then like more for a single share, like a single share of the

Chris Gammell: NPW. It doesn't get you more chips, but it gets you the same number of chips at a much cheaper price. So I got a hundred parts and it costs the research group. It didn't cost me. Uh, it costs about three and a half grand to the university. If you were to do it as a unit, as an individual, I think you, it's a bit more expensive. It's maybe, uh, maybe four, four to $5,000, but it's, it's in the realm of, you know, if you're trying to do a startup and have a prototype, it's in the realms of possibility.

Anthony Wall: I mean, 50 bucks a chip these days, that's, that's standard, standard fare, right? Yeah. Given the shortages.

Chris Gammell: Yeah.

Anthony Wall: So I'll build my own STM 32. Damn it. Damn right.

Chris Gammell: Yeah. Just, you know, buy an arm license and go hell for leather. Right. Yeah. Right. Right. Right. So, so yeah, we, we came up with the same kind of issues in terms of, uh, sourcing chips from our end. We send away our files to, to Euro practice and they make the masks and they send it off to TSMC and, uh, you know, TSMC disappear for a few months and we're, we're given a date in August. We tape out in April and we're given a date in August. You'll have your chips in your hand by then. Uh, you know, August rolls around and TSMC say, oh, you know, uh, we'll have them to you in a month. And, uh, you know, September rolls around. Oh yeah, we'll, we'll have them to you in a month. And, you know, in the end, like, you're kind of just like, do my chips exist? Like, did I spend the last year just like, you know, having the crack and making some, some nice, some nice, uh, abstract art, you know?

Anthony Wall: Oh yeah. Do you print that out? Do you mean, do you, do you actually hang it on the wall? I would imagine that would be, I guess it depends on which, uh, which layer you want to print out, but.

Chris Gammell: I, I have it as my desktop on my laptop. Uh, that's, that's, that's, that, yeah. I'm trying to figure it.

Anthony Wall: Can you send it to me so I can pretend I designed one too? Yeah, absolutely. Yeah. Yeah.

Chris Gammell: It's, it's about a, it's about a, uh, you know, a 20 megabyte image. I've kind of like, I assume it must be, you know, breaching. Some NDAs like now, if you look at my screen, you're, uh, you're now breaching an NDA, I'm sure. Right.

Anthony Wall: Right. Yeah. All you have to do is memorize it.

Chris Gammell: Yeah, exactly. Yeah. You know, just, uh, just gobble it up. Turn it right.

Anthony Wall: Yeah. I forgot.

Chris Gammell: Yeah. So, so, so long story short, we got our chips back kind of mid October, November when we should have got them in, in August. And I mean that, you know, it's not, not too bad in terms of the lead times, you know, you're talking about when you go on DigiKey, but, um, you know, when you're, when you're aiming for conferences and stuff, it's, uh, yeah, I was going to say, if you're trying to graduate

Anthony Wall: that, that might, uh, that might put a kink in the hose, you know?

Chris Gammell: Right. Yeah. Yeah, exactly. So, so when you get the, you get, you get some, you know, tiny little pieces of silicon back, you need to be able to use them. So, uh, you need to put them in a package. Um, so your practice will also offer you some, some packages. Uh, so you, you tell them how to bond the package out, what pins you want to go, you know, what pads on the chip you want to go to, what pins and, and they'll, they'll package that up and yeah, you design a test PCB and hopefully everything turns on. It's interesting actually. Wait, wait, wait, wait. Don't leave us hanging. Did it? Yes. Somehow it didn't, you know, the, the fear of like the first time you powered on, you're slowly ramping up the current limit on the power supply. I can't even imagine.

Anthony Wall: Yeah.

Chris Gammell: Yeah. No, it worked somehow. Somehow it worked. Um, yeah. Yeah. I was incredibly lucky to, to, to say the least.

Anthony Wall: Surely. I mean, does, does your, like, does your, your boss, your PI, whoever, whoever it is, they sit you down and they're like, let me tell you a story about a time when the person's chip didn't work. I mean, like you must get like stories like that. Right. I mean, what happens?

Chris Gammell: Yeah. I mean, I I'm lucky that, you know, the, the environment in Tyndall is very kind of, it, it's a friendly research environment. It's a positive environment. So, you know, they're not like, you know, you're fired if this doesn't work, but there is the stories. Like, I mean, you know, it doesn't turn on. Like the worst case scenario is your chip comes back and you power it on and it's up in a puff of smoke. You have no way of knowing what's wrong with that. Like you can go back on simulations and you may find kind of hints, but if your chip goes up in a puff of smoke, as soon as you turn it on, like there's no troubleshooting there really. You know, it's no. Yeah. Right. It's, it's another six months of your life. You, you go back and you, you figure it out. You try again.

Anthony Wall: 99 more chips on the bench, 99 more chips on bench. Yeah.

Chris Gammell: Oh boy. Yeah. No, I, I, I was very lucky. I only, I only popped a few of them and you know, I gave them a happy burial and you know, you know, we, we, we got very, very lucky.

Anthony Wall: I mean, really a chip going puff is kind of its own Viking funeral, you know? Exactly. Yeah. Yeah.

Chris Gammell: It's going down. But you know, the tools these days, and that was another thing I was kind of, I said, I'd mentioned to you, you know, you, you've spent a lot of your time on the, on the PCB side of things. And I've spent most of my time on the IC side of things, you know, I needed to design a test board, test this thing. So I made that kind of reverse transition from IC back to PCB. I had never done a PCB before, before this. So it was kind of an interesting, like. Harsh introduction. Yeah. Yeah. Very much so. You know, like 20 years ago, well, realistically 30 or 40 years ago now, chips were made, you know, chips were laid out using layout tape on, on these big boards and, you know, tape out literally meant, you know, the layout tape is going out to the fab. Right. Right. So, so, but nowadays it's really difficult in the IC world to make a critical mistake. You can make functional mistakes and you can make, you know, your architecture, your bare idea may be, uh, may be flawed, but like, it's difficult to have a short circuit, for example. Right. So, you know, when you're designing a PCB, you have millions of components to choose from. You can download a footprint that, that Johnny made a few years ago that you have no idea if this footprint is good or not. Right. You can download a footprint from whoever, easy EDA, you know, the IC design world, like I have maybe a hundred different types of transistor. I can change the size and the dimensions of them, but I have only that number of transistors that I can play with. And I know they all work.

Anthony Wall: And, and those come, is it, that comes from the fab? Is that right? Or that's like the design package that comes from the fab?

Chris Gammell: Exactly. So you have what's called a P cell or a parameterized cell. So if I want to put down an N channel MOSFET, for example, I'll take a, a parameterized cell of an N channel MOSFET. And that let me, you know, change the width and the length of that device. But I can't like change the, you know, the stack up, for example, I can't change the doping or any of that sort of crack. That's, you know, that, that's relatively fixed at our level.

Anthony Wall: Oh, interesting.

Chris Gammell: Right.

Anthony Wall: Wait, so do some chips actually change the doping or, or is it just like.

Chris Gammell: You can like, I mean, so, so when, when TSMC gives you a PDK, the understanding is that firstly, you know, when TSMC gives you a PDK, really it's for the digital guys and the analog guys, you know, they can play with the transistors too, but this is for the digital guys, right? So, right. So they'll give you a number of flavors of transistor. So you might have a low threshold voltage transistor, a regular threshold voltage transistor and a high threshold voltage transistor. And what that's really far is for the digital logic to consume, you know, less power and go slower or more power and go faster. It's to kind of, you know, right. It's to tweak your digital logic, but obviously that has analog, you know, the analog guys can use that to create slightly different effects too. Right. You'll be given different flavors of these parameterized cells and you need to, you know, decide which threshold voltage I want. And then I need to decide how big do I want the transistor? And they're the knobs that you really have. Got it. Yeah. Oh yeah. Yeah. Okay. Yes. So, you know, the vast majority of time.

Anthony Wall: I'm just thinking about my, again, like my knowledge is about 15 years out of date now, but what I do remember is that at different stages throughout the semiconductor process, there is a mask layer that's all very, very thick mask layer that also goes towards the implantation steps and like, you know, different parts of the circuit are going to get different levels of doping and things like that. And so someone's doing that, but it sounds like it's, that piece is already set. In the PDK or pretty much set if, if nothing else like you. So you're saying that it's very tough to actually then go and open up part of a mask so that it gets more doping than, you know, like in a, in a wide swath of area, even in analog designs.

Chris Gammell: Yeah. Like, I mean, you can write. Okay. So you can't say, for example, mess with the, I can't point to the particular area and say, I want this level of doping and tweak it like an analog knob. I can't do that. But technically if I wanted to, I could go into the layout and I can open up a, an N diffusion or a P, you know, an N well or a P diffusion or something or a piece of polysilicon. I can put that, all of those things down on the layout. And in certain instances you do that.

Anthony Wall: Okay. So that, would that be the equivalent of like going in and like modifying Gerber's kind of, is it like at that level of like kind of wild westiness on, on that aspect of it?

Chris Gammell: Yeah. And I mean, the tools will tell you if it's manufacturable, but, and, and, you know, there are, there are tools that will tell you, for example, if it will cause a latch up or an ESD, you know, or latch up.

Anthony Wall: Oh, wow. Yeah. Okay. That's pretty cool.

Chris Gammell: It is. And that was what I noticed really in terms of going from IC to PCB is like, if you do all the DRC checks and you have to do the DRC checks, it will tell you categorically, this chip is 90% manufacturable, right? We're 90% sure this is manufacturable and you'll get something back that, that is what you sent away. And you do the LBS checks, layout versus schematic checks. And, and, you know, you've a relatively, you're relatively secure in the knowledge that this thing is what you want. In a PCB, like it was really scary. Like, you know, I.

Anthony Wall: Lower stakes. I just to start with, I think that's, and I think that's why the tools aren't quite the same, but yeah. Lower stakes. Yes. Well, I mean, my PCBs were the same price as my chips.

Chris Gammell: So when you put it that way, it's, it's wild, right? I mean, you know. Yeah. But the turn time is a lot more stringent.

Anthony Wall: Yes. Correct. Right. Absolutely. Yeah. Yeah.

Chris Gammell: Yeah. It's, it's a time. Yeah. It's a time cost as opposed to the monetary cost. But like, you know, we, we, we, we have a guy who does the PCBs in, in our, you know, we, we, it, there's a guy, right. You know what I mean? And I remember emailing him being like, Hey, you know, could you send me the stack up? You know, what are the dielectric constants? And you know, what's, I was trying to do some matched impedance traces and stuff. And you know, what are the minimum tolerances on the, you know, like widths and spaces and stuff. And he goes like, Oh yeah. It's a four layer stack up. Yeah. Take it to no four point, whatever for your dielectric constant. You should be grand. And in my head, I'm going mental. Like, I mean, there's like, I don't know how many thousand DRCs that I just cleared on my chip. And this guy is telling me to take the dielectric constant, you know, four-ish, you know. Four-ish, yeah. Yeah. That's about right. Yeah. So that was.

Anthony Wall: I mean, 10% is usually okay for, you know, matching these stuff, right? Yeah. Pretty standard.

Chris Gammell: Absolutely. It was, it was the fear factor of like, I could do something in my EDA tool. I could design a PCB and no one is going to tell me that this thing can't be manufactured. I mean, like I might get an email from the fab saying, what are you doing?

Anthony Wall: Yeah. Right. Right. Ground is connected to the power rail or the power layer layer. That's bad.

Chris Gammell: Well, I mean, they may not even tell me that. Like they may tell me, Oh, your tracks are too narrow and you may have like a, you know, you may have a yield issue, but they don't necessarily know that my power is connected to ground. Maybe, maybe that was a good point.

Anthony Wall: Yeah. Yeah. They don't really have that. Yeah. That's true. Sometimes, sometimes they'll tell you, but not, not always. Yeah.

Chris Gammell: Yeah. And like, I'm not sure, I'm not saying it's a deficiency in the PCB design. Like I've got maybe, you know, 20 components that I can choose from. You've got millions, right? There's no way of testing functionality on a PCB really, unless you've got like a simulation library of every component, you know.

Anthony Wall: Is that because the, the, the CAD software for chips is more integrated from the layout stage, the schematic stage, the layout stage that is just inherently more connected? Or is it because?

Chris Gammell: I think so. I think so. Like, I mean, there are, you know, there's, there's the cadence tool flow, like in terms of what, what people use and what people pay for largely there's the cadence tool flow and there's the synopsis tool flow. So if you're a foundry and your tools, your PDK doesn't support cadence or synopsis, like nobody's designing you. It's not possible to design that chip. Duopoly. Right. Exactly. So it's in their best interests to make sure that, you know, they're, they're very, very well integrated. And I think that's changing. And I certainly hope so in terms of, you know, that the open source IC design movement. Oh yeah. Really, really exciting.

Anthony Wall: Maybe it's making a little movement, but I think it's going to be a very small dent for a little while yet. So yeah.

Chris Gammell: And it's, you know, it's a, it's got, it's, it's got, you know, there are certain things that will be good at and it won't be the bleeding edge, but I think it has a place and I think it should be certainly explored. Like, you know, 20 years ago, 30 years ago, the idea of, you know, of KiCad would probably have been kind of laughed at. Of course you need to pay for your PCB design tools. You know, I don't know. How long has KiCad been going actually? Yeah.

Anthony Wall: Yeah. About 30, 30 years. Yeah. It was like 91, 92. So.

Chris Gammell: Okay.

Anthony Wall: Right. Okay. But it was definitely laughed at it then at that point, I think.

Chris Gammell: Right. And I mean, people are taking it seriously now, I think. Same with software. I mean, software, you know, has grown into this open source ecosystem. I know hardware is a lot more difficult, but it would be nice to see, you know, more, more open source effort.

Anthony Wall: Yeah. Totally. Yeah. I think some of it, you know, some of the stuff you've already talked about here with like the synopsis and cadence, it's like, it's, I think it's almost vertically, not by design, but like almost by necessity, like the, the sub, you know, the small number of players in the market and the incredible complexity is just like, I don't, I honestly don't know how you'd get around. And I think some of the information would actually be lost. So just thinking about like, I don't ever send my schematic to a board house. Right. I just send Gerber's. It's just very generic, very old format, you know, and it's just like, all right, just make this thing. I said, how to make it. And so there's really none of that knowledge of, oh, well, this component is connecting these two things together. Okay, fine. They don't care. I don't really care. It's, you know, low stakes, like I said, but like in a chip fab, like I'm guessing there's a little bit more knowledge of, you know, the tying back of the PDK to like what components are actually in there. And yeah, like what, what happens when you send out an analog design to a, you know, a TSMC to someone there looking at it, or is it just like, they're like, oh, well, here's the files. We'll put them into our masks.

Chris Gammell: Right. So, so before you send it away, they have like very, very extensive DRC checks. I mean, like, you know, they'll check the layers, they'll check the, the number of elements in a polygon, it's just stuff like you cannot get away with anything. So, yeah. So when you're sending that. The stakes are so high. It just seems like. Yeah. And also not only your, so if you're doing this in a multi-project wafer with other people's stuff, right? If your chip is causing an issue, you'll affect the chips around you. Right. So a prime example of that is density. So, you know, when you're making a chip, you deposit a metal layer and you polish it. Uh, you, you kind of, you, you make, you make the surface plain and then you add another metal layer. If you have, for example, too much metal or too little metal, more importantly, in your chip, you'll create a little indent when they polish it, right? Yours will polish more than everyone else's. So the DRC checks need to check if you've got a high enough density of metal that the polishing will, will be even enough. So you don't mess up both your own chip and someone else's. Do you put in like a, like fake metal structures then if you don't have enough density? Exactly. Yeah. So we have a script that will, that will look at each kind of zone sequentially and kind of fill it in with little polygons of metal here and there. Exactly.

Anthony Wall: Come on. You don't put in like designs, you don't sign your name and sign your name in there for the logo.

Chris Gammell: Oh no, no, no, no, no, no.

Anthony Wall: Like a graffiti tag.

Chris Gammell: So, so I'm, I'm known in our office for like, no matter what month of the year, like, look, we have one season in Ireland, I always wear shorts for that one season. So year round, I'm always wearing shorts. So, you know, I said, look, I, you know, there's always going to be at least one short on a chip. So I put my, I put like a 10 micron pair of shorts on my chip as one of the dummy metals. Fantastic. Yeah. And I mean, that's, that's also necessary that like, it's funny, but you do need some kind of locating factors. Modern chips have so many metal layers and so much, so much of this kind of density fill that actually, when someone goes to bond the chip, it's really hard to see what way is up. What's pin one here. Yeah. Yeah, totally. That, that is one of the requirements, you know, that your practice have is that you need a logo or you need something, you know, something funny to, to, to locate. Yeah. Orientation element. Yes. May also be a funny logo. Absolutely. And you know, I, I, I didn't tell my PI, you know, until, until after I, I sent in the GDS. Oh yeah, by the way, there's a pair of shorts on this chip. Yeah. That's great. That's great. But anyway, so, so you, you do all these DRC checks and then someone in Europe practice will also run their DRC deck. Now, ideally your DRC decks are the same. And if you're clean, it'll be clean. Once it gets to Europe practice, we all know what happens when DRC decks are sent back and forth and emails and stuff. You take different versions and stuff like this. Oh yeah. Yeah. We talked to Michael about that recently. Right. Yeah.

Anthony Wall: Picking the wrong zip file. Yep.

Chris Gammell: Exactly. Exactly. Right. So, so, so sometimes you'll have DRC issues kind of that way, but yeah, you're relatively certain sending it out that it's manufacturable and it won't mess up other people's stuff. Yeah. Yeah. Now, and the fab, I should also say like the fab don't care, doesn't care what, what you're doing. They don't see your schematic. They just care that it's. Okay. So it is similar in that way. It is similar in that way. We have quite extensive LVS layout versus schematic. And this is, you know, as soon as you turn on your, so you do your layout, your chip is looking well, you run your DRCs, your DRC clean, happy days, right? You turn on layout versus schematic, and it basically checks that the transistors that you've drawn in the layout and a transistor is defined by a gate and, you know, two, two diffusions, either side of it. Right. So that's what a transistor is. And the LVS will pick up if it sees a gate and a diffusion, either side of it, that's a transistor. Right. So if that, obviously the number of transistors in your schematic must equal the number in your, in your layout. And here is the kicker, right? A transistor is a four terminal device. You learn that as soon as you turn on the LVS, right? For the majority of your design. Body, body, body. Yeah, exactly. Right. So the majority of your design, oh yeah, I'll just tie the, tie the bulk to ground, da, da, da. But you forget that, you know, the chip is a whole block of silicon. And unless you do something fancy, all those bulks are tied together. And if you've got different power domains, for example, they'll, they'll show up as a, what's called a soft connect, a connect through the bulk silicon. And that will fail your LVS because you could have latch up, for example, through different power domains, you know?

Anthony Wall: You know, and, and they teach, they teach like the four terminal device thing in school, but boy, that didn't make any sense to me. You know? And it's just like, I don't know, it feels like one of those mental models where they should be like, okay, it's four. We're going to talk about three, you know? Like, and I know that's what eventually happens anyways, but like, until you get to the level that you're talking about and actually designing a chip, like it's so hard to visual, it's still, I mean, it's still hard to visualize, honestly. And it's just, it's.

Chris Gammell: Yeah. And, you know, I think teaching transistors and the idea of, you know, it being a, you know, a valve that you turn on and off and depending on how much you turn the gate on or off, you, you value, you vary the current. I think that's what people need to know. And then when you need to design a chip, fine, learn about the bulk, but you know, it's okay not to know about it until you need to know about it, I suppose.

Anthony Wall: Yeah. I mean, there are effects, but it's just like, I don't know. It's just so flood. It's like, you have to kind of like really tunnel vision down just to get the stuff that's actually really tough in the first place. And then, and then like, yeah, okay, you're going to zoom out at some point and we'll take those blinders off and be like, oh crap, there's a lot more to learn, but that's also the fun part. Right. I mean, yeah, exactly.

Chris Gammell: You know, an, an engineering course is, you know, a finite amount of time and you can only learn so much, you know, certain things you need to cut out and certain things you don't need to know. And, you know, you inspire people to learn these things themselves at a later date, but yeah.

Anthony Wall: Yeah. So one thing I always wonder about with this sort of thing is how much, so you mentioned like you, you know, you kind of had this a hundred transitions you can pick from, but then are you constructing your own like elements that then you kind of bundle together more? So like, uh, like do you have your own current source that you might drop at multiple places throughout the design? Yeah. And that's like a component quote unquote.

Chris Gammell: Absolutely. So, so yeah, chip design is very hierarchical. So, so when you boot up cadence, what do you see? You see, uh, you see your work divided in, in three columns, right? It's, it's similar. I must admit, right. I'm, I'm, I'm intending to transition to key CAD, but I haven't yet in Eagle.

Anthony Wall: You've got no confessions are required on this program. So it's okay. I forgive you, my son.

Chris Gammell: I'm trying to, I'm trying to open source, if I find my life. So, so key CAD is next on the list, right? I, yeah, it's going to take a while, but in the chip industry, I think so. Yeah. Yeah. So, so anyway, right. You, you open up cadence, right? You have a library of components, you have a cell and you have got what are called cell views, right? So, so your library might be, I might have like, for example, a library of op amps, a library of different, uh, ring oscillator delay cells. I might have different bias, a library full of different bias circuits, stuff like this. Right. And then cell views will be the particulars. And then, you know, uh, uh, cells will be the particular, you know, uh, a 10 micron bias circuit or a 20 micron bias circuit. And then the cell views will be whether it's a schematic or a layout. Right. And when I have a top level of a chip, it's effectively like a set of Russian dolls made up of these different components kind of layered on top of each other. So, so yeah, it's, it's, it's incredibly hierarchical and you may have, you know, dozens of layers, you know, I may have a cell that's 20 transistors big and I may put that around different places and have it on multiple hierarchies and, and stuff like that. So yeah, it's very building blocks oriented. So you can reuse things hopefully. Got it. Yeah.

Anthony Wall: Well, and so, I mean, so in our conversation before the show too, like, so this is a mixed signal I see. So there's also not just analog, right? So it's also got other stuff in there. So part of what I was working towards is like, is there that everything else that has to tie into it? So like all the control circuitry, things like that, like, is that something that you get to just be like, knock on your neighbor's cube and be like, Hey, can I use your digital control circuit to turn on this other aspect? Or how does that work?

Chris Gammell: Ideally? Yes. In my case, no. And, you know, that's, that's a, you know, it's a pain when you're taping out, but afterwards it's great because you kind of gain a kind of a glancing understanding of what everything does. So, yeah. Right. Yeah. So, so I, we have, we have a digital person in our group and, and she's really, really helpful in terms of doing the digital stuff, but she kind of says like, look, you know, figure it out when you get errors, come to me kind of a thing. So yeah, I mean, you, it's, it's very, you know, it's very log. The digital is written in very log, right? So let me, ah, got it. Okay. So, so let me put context on it. Right. I've got a ring oscillator and it's giving me bits out at one gig sample per second. Now there's not a hope in hell. I'm going to be able to get that off my chip easily. Right. The pads are far too slow. The bombed wires have inductance that would stop me getting that off too easily. Right. So I needed to downsample the output of my ring oscillator. So, so you can use what's called a sync filter or a decimation filter to do that. And effectively what that does is effectively it filters the data first and then it downsamples it to, you know, you might, instead of having one bit at one gig sample per second, you might have eight bits at 32 meg samples per second, for example. Right. So, so you, you program that in Verilog and you can, you know, you have an ideal Verilog test bench, like you would just simulate any, any Verilog, you know, whether it be for FPGA or whatever else, and then you synthesize it. And what synthesis means is it fits your Verilog. And I mean, it's the same with an FPGA to available blocks, but the available blocks are what are known as standard cells. So you might have an and, you know, a four element and a two element and you had a flip flop, different flip flops. So, so you synthesize the Verilog, which creates a gate level Verilog. And then you, you, you then do place and route. So you actually place the cells in the chip and wire it. And, you know, it has to be timing. So you, you know, the tools can handle all of the actual putting the digital on the chip. You just write the Verilog. And I apologize to any digital designers. I know it's a lot more complicated than that. But for, for my simple case, that's, that's, that's how it works.

Anthony Wall: So, so then, okay. So then all that stuff's done. And then now you basically have kind of the equivalent gate level. It's just using the standard transitions, like you mentioned. Right. So now it's like this, what I'm really imagining in my mind is like, you've got this pastoral, nice, well laid out farmland. And then you're basically like, there's like a hard, like county line. And then you're like going into like a downtown and there's just like a cityscape of like, of like digital mess that you, that you're, that you're basically going into Lincoln tunnel. And like, you know, like all of the data is going through there or something.

Chris Gammell: Yeah. Yeah. You know, I'll, I'll send you a picture of the chip. Like, I mean, there's like, you know, uh, my, my little analog section, I spent all my time, like, you know, perfecting and tweaking my little analog section. And then there's just this big behemoth of digital that I made like in a panic at the last minute. And it's just ugly. And it's right in the middle of the chip. Right. Yeah. Yeah. You have to, yeah. Yeah. And I, I mean, you have, it's funny to look at, but you have to be careful. You know, there, there's a finite, you can get a finite amount of current through the power supplies in your chip. So if your digital takes a big gulp of current, uh, you need to be careful that that doesn't take away from the available supply on the analog side. So you'll see there's like a slit right down the middle of my chip. So the ring of pads is split right in the middle and the right hand side is all the digital stuff. And that's, that's dirty. And on the left hand side is the analog stuff. And hopefully that's quiet.

Anthony Wall: And how do you actually separate those domains? Cause you don't get to say, oh, actually, no, this is a different wafer. Same wafer.

Chris Gammell: Yeah. Right. So, and I, I mean, the point I made earlier about, you know, the bulk you get, you get a piece of silicon, the bulks are connected. Um, you can do different tricks. So you can do what's called a deep N well, which effectively is a buried. So, so your, your silicon wafer is likely P type, but what you can do is you can, you know, blast, uh, layers in and you can have a buried layer of N type. So you have a kind of a NP, sorry, a P N P sort of stack vertically, and you can kind of stitch around that. So you make a tub like a bathtub. It's kind of isolated because you've got P reverse bias P N junctions. So you can put your sensitive analog stuff in this, uh, deep N well. And therefore it's technically, you know, it's on the same chip, but it's technically at least electrically isolated or capacitively isolated from the, the, the digital stuff and the digital stuff just sits in the tub or sits in the main, uh, substrate. The digital stuff just doesn't care. Right. And you have, you know, you have separate power domains. You, you make sure that you have a digital supply and then you have a clean analog supply on the chip and stuff like that too.

Anthony Wall: So on that, the deep N well thing. So is that like the early in the pro I'm just, again, thinking back to my limited early semiconductor days and like, I remember there'd be like some early stages where they're just like doing high energy. Like you go to send something to implant with like these huge open windows where the, the mask is really open and they're just like cranking the thing all the way up and just trying to shove some kind of doping as deep as they can into the wafer. Is that like the deep N well type of thing?

Chris Gammell: Yeah. That I, I, unfortunately I would love to get to make this myself. I don't. And you know, TSMC certainly aren't going to tell me how they do it. But again, you know, if they want to, they're more than welcome to, but they're not going to tell me how it's done. They're a little close to the vest on stuff. Yeah. Right. I know, for example, like making, uh, if you want to make a good drain, you know, it's, it's difficult to make it a good like drain and source. So I know you really need to beat the dopants in there again. You know, I'm not a device guy. I don't know, but yeah, it's that idea. I can't remember how they do the deep N well, the buried layer. It's some weird ion implantation thing anyway. Yeah.

Anthony Wall: Yeah. Cause I remember some of the stuff they, I mean, like some of the stuff they actually like wind up, they have like a cyclotron effectively and they just wind it all up and then they shoot it out, out the back end of this cannon and it shoots dopants. It's insane how it works, but it like shoots dopants into a wafers. Okay, cool. And then sometimes they actually bake it in with like a, right. An oven. Right. Yeah.

Chris Gammell: You have to repair the damage done by that blast. And I know you can do funny things with the energy. Like if you, if you have a certain energy, you know, it'll go a certain depth, you know, that's right. Yeah. Yeah.

Anthony Wall: Killa electron volts, mega electron volts, all that stuff. It's like, yeah, they talk about the, that sort of thing.

Chris Gammell: Yeah. You know, it's the real engineering.

Anthony Wall: It's crazy how it works. I mean, like, yeah, it's just like, that's how, I mean, that's how we're talking right now across vast distances. Right.

Chris Gammell: Yeah. It's crazy that it works. Yeah, exactly. You know? Yeah. Yeah. Yeah. It's done on a very large scale. Yeah. Like, yeah. A hundred years ago, you were, you were two weeks away. You're now 70 milliseconds away because of this sort of stuff. It's incredible. Right. Yeah. Yeah. Yeah.

Anthony Wall: Well, okay. So you got the ship working. You're hopefully on your path towards being doctor. What, what's, what's next? I mean, jumping to the chip company that sponsored you. Is that, is that the path that you mentioned is your path or.

Chris Gammell: Yeah. I mean, that's something else. Something I'm trying to figure out at the moment. Like it's, you know, the problem with this, right. So firstly, we, we talked before the show about, you know, what, what, you know, people who want to maybe think about going down the same route as me or a similar route. One, one piece of advice that my, uh, one of my co-supervisors gave me when, before I started, before I signed the piece of paper was look, a PhD is like a religion, right? It means nothing to anyone else apart from you, right? They don't care that you're doing one, right? They think, oh, cool. You know, you get a, you get a new funny name. No one cares, right? That's right. You, you're, you know, you're not entitled to any more money or any more respect. You do it because you believe in it. And whether you choose, it's then a tool in your toolbox, whether you choose to use it or not is again, you know, it's up to you, you know, that, that, that was kind of profound. And that's kind of the juncture I'm at right now. Like, I mean, I want to stay technical. I, I couldn't be doing all of this outlook all day sort of crack, but you know, so, so, you know, the, the managerial route isn't, isn't probably where, where I'd like to go. Uh, my options are, you know, go in, go into the company that, that sponsors me. And, you know, I, I do interesting work. Like, I mean, I, I'm a designer, but the trade-off is, you know, you have security, but you're a small part, you're a small cog in the chain effectively. And you're doing a small, you're, you're doing a small part of a much bigger operation.

Anthony Wall: Yeah.

Chris Gammell: Do you go into something?

Anthony Wall: Right. Are you doing just current, current starved ring oscillators for the rest of your career? Well, that might not be as, as rewarding as something else, but you might get really deep and find some really interesting things in there and become a world expert. Right.

Chris Gammell: Sure. Yeah. And I mean, there's benefits to being, you know, the guy in the company, you know, when you grow out your beard and you're, yeah, you know, right. You know, you're, you're, you're, you're, I can't wear sandals to work with socks on.

Anthony Wall: I'm the guy.

Chris Gammell: Yeah. Yeah. Exactly.

Anthony Wall: I'm the guy.

Chris Gammell: Yeah. I like, look, ideally, you know, what would I love? I, I look at the, the software people with envy sometimes, you know, if you have a computer and you have the ability for a few months not to get paid, there's a chance you might stumble across, you know, you know, a good software startup, right? It is tougher in the chip world. What would it take for me to, to, to, to, to do a silicon startup? Well, obviously first a good idea, but then you have the, the logistical stuff. How do I get CAD tools? How do I get TSMC to give me a copy of their PDK? How do I get the money for my first prototype run? It's tougher, but. You know, again, look, if anyone wants ring oscillators, I'll gladly, uh, gladly set up a startup and we can, we can, we can do business. Right.

Anthony Wall: I mean, that's the thing too. I think that the, the existing stuff you've been working on, that stuff that you are, I mean, I said world expert before, but like, you know, PhDs are supposed to be here that world expert in that very narrow PhD way that PhDs are right. But there's a lot of stuff you need to put around it as well. Right. And so there's just other business aspects, other chip stuff, you know, how, how, how much do people need just that one thing? And so.

Chris Gammell: Yeah, exactly. Yeah. There's, there's a lot more to it than just what one person can, can provide to make a chip. You know, we love, we love playing with things, but you need the people to do the marketing and you need the people to, to, to do the finances and all of that sort of stuff. You know, it's, it's very difficult. Like to go from, I have a cool idea of how to make a ring oscillator better, uh, to actually here are my chips on DigiKey. It's, that's a long road and that's not a one person road either, you know?

Anthony Wall: No, that's right. Yeah. I think that's right. I think, yeah, you have to have either like a non-technical co-founder that you trust implicitly to, you know, run the, run the business side. And like, if you want to stay technical and I think that, I think the benefit of these days with startups, especially in the chip space is like one. I mean, well, we've seen in the news that every single chip company gets gobbled up these days and I'm sure that's not a trend that's going away anytime soon. Absolutely.

Chris Gammell: Yeah.

Anthony Wall: I think access to fabs is more than it was, right? So like fabulous semiconductor startups are a thing now, maybe not a common thing, but they are possible. Oh, certainly. So that's good too. I think you could make a, a case for even adding on risk five style processors along, you know, your digital section might be a, not needing to talk to arm for their licensing and things like that. So I think in all these ways, there's a lot of benefit there, but then are you talking the right niche? Are you, you know, like, does the world need that current measurement device that you're talking about, whatever. So yeah, it's just, yeah. And then, and then dealing with all the funding stuff is just a lot of.

Chris Gammell: Yeah. Like the tooling is coming, right? We were talking about the open source tooling. That's coming. I think the fabs are, are sent starting to sense what direction the wind is blowing. You know, you've got the Skywater fab with their open source PDK and, you know, the funding is there. Like, you know, I'm lucky enough to be in a research environment. And if I, if I put a proposal together, I, you know, I could do a postdoc and kind of work on a startup. Like that's, it's possible and it's becoming more possible at the moment.

Anthony Wall: Yeah. That's good. That's good. There's options.

Chris Gammell: Absolutely.

Anthony Wall: Yeah. And I think the only thing that's more glamorous than the life of a PhD student is the life of a postdoc.

Chris Gammell: Oh yeah. Like, that's great. Like, you know, you, you get paid nothing for, for your postdoc and you don't even get a fancy new name.

Anthony Wall: Like, who's that broken person over there? Oh, well, they're the postdoc. Yeah. You can just ask that.

Chris Gammell: Sorry, any postdocs are listening. I don't mean to, you know. No, like I would, I would like it, I would love to be a postdoc, right? I would just love to get paid for the privilege of doing so. You know, my ideal, you know, if you could get, you know, get, get paid, like you'd get paid in a company and get, get to sit and like solve people's problems. Like have PhD students come to you and just work through problems with them. Like, come on, that's, that's a dream. That is cool.

Anthony Wall: Yeah, totally. I think it's called consulting. You know, you should just become a consultant. Yeah.

Chris Gammell: If anyone needs ring oscillators or is trying to figure out how they work. Yeah. No, that right. I like, yeah, no, Chris, I'll be on the forum right after this, you know? Okay. Great.

Anthony Wall: Bring it on. Bring it on.

Chris Gammell: Yeah. But I mean, yeah, the, you know, the, the, the tool chain is, is, is getting better for, for, for the, one other thing I wanted to, to kind of briefly mention was, you know, we've, we've had a lot of change recently and, and the way we learn, for example, circuits is changing and has changed. Like, right. I'm teaching half a course on nonlinear circuits this year. And the change that has occurred in, so I took that course in 2016 and the change that has occurred in those short few years is, is great. Right. So back then, and it was the same for you, I'm sure. Right. You go into your circuits class and you blast equations. You write frantically for an hour and you have equations, you know, you have the lecturers who make you take everything down. Right. And you, you go out and your mind is frazzled and you have taken very little in. And now the pandemic has, has, has enabled us to say, right, you know, okay, first of all, maybe our lectures will be in person or maybe things will change again and they will not be in person. So we need to have the facility to ensure that we can teach our students either way. So, so what's kind of happened is, you know, we give out the notes, first of all, so students don't have to frantically take things down. And furthermore, they don't have to, you know, they can pause a lecture. The idea that a lecture is recorded and they can pause it, I think is like, you know, yeah, big change. And yeah, absolutely.

Anthony Wall: So you can, I mean, like, I just, I, I actually, that visceral, like a visceral feeling of just like the, the, I, I remember this, like sitting in the, in the auditorium or wherever the class was, bigger classes were usually more stressful for me, but like sitting in the auditorium, looking around and everyone else is just nodding their head. And I'm just like, what the hell did he just say? Like, does nobody going to stop nodding their head for a second? And then like, also like the, I don't want to raise my hand. I don't want to stop people, like whatever, you know? Yeah. And it's just, yeah. And it just keeps rolling, you know?

Chris Gammell: Yeah. And I mean, if you, if, if you stop to think about it for a second, you, you've lost, right? You, you, you don't have that equation taken down. So it's now, now not in your notes. Whereas now, like, it's great. We are able to say, right, here are all your notes. Don't worry about taking it down. Sit back. Right. Look, put your phone away.

Anthony Wall: Let's put the brain card in motion, right?

Chris Gammell: Exactly. Right. And what we do now is, you know, we, we'll, we'll go through the slides and then we'll take a second and we'll open up. I don't know if you're familiar with Falstad, the circuit simulator.

Anthony Wall: Oh, I love Falstad.

Chris Gammell: Yeah. Right.

Anthony Wall: Awesome.

Chris Gammell: Yeah, exactly. And you could say like, look, right. Look, I've given you all the equations for a BJT, a bipolar transistor, right? Look, fine. Right. Go back and study them for the exam. Look, if I put a little bit of current into the base, a load of current is going to come out of the emitter, right? You know, yeah.

Anthony Wall: And the visualization piece of that is super important too, because it's, you need to, you're literally connecting the dots, literally watching someone connect the dots in that case, right? As you're doing that, because the dots are flowing, you're like, oh, okay. Visualizing current. Yeah. Whatever. I just, yeah, that is super important. Yeah, absolutely. Is this just the way you teach it or is everybody doing this? Because I could see it also just being, you know, some people just doing the de facto and blasting through stuff.

Chris Gammell: There's a kind of a debate right now. And I know different people on different sides of this argument about like, you know, should you teach the, so you're allotted two hours per week to teach your subject. Should you teach the subject in those two hours? Or should you give the material to the students beforehand in video lectures and then have like a tutorial for those two hours, right? I'm firmly in the, have the tutorial for the two hours camp.

Anthony Wall: Yeah.

Chris Gammell: Yeah. Because I think we do something quite practical. I think it's, and I know, you know, a good number of, of, of academics are going that way. And I, I think power to them, you know, absolutely. The other people kind of say, well, you know, maybe the students won't look at the video lectures or can we implicitly trust them? And then I said, look, college is expensive. We motivate this industry. Yeah. Yeah. I, I believe that, you know, people go to college, hopefully to learn to have fun too, but hopefully to learn, you know, let's trust students and give them the tools and see if they come.

Anthony Wall: Yeah. It's going to bite them in the ass at some point if they don't, you know, if you don't actually learn it, there's no getting around it sitting in your cube at Intel or wherever you end up, you know.

Chris Gammell: In Ireland, we have fairly grueling exams. So, I mean, look, you're, you have a gate that you, if you don't keep up, there is a gate that you can't get through. So, you know, it's in your best interests to keep, keep up. Right. But yeah, I think, I mean, we've had too many generations of people who understand the equations behind the circuits that we do or, you know, know the equations, but don't like that. That we're never given the, the intuitive grasp of like, oh, in, in your head, if you can twist a knob to see how, you know, how this, how this changes the parameter in, in the circuit or whatever, you know, I think.

Anthony Wall: I mean, at this point, I'm guessing people are going to start having falsehood style schematics and little dots in their head, you know, I internalize tools like that all the time.

Chris Gammell: Right. Yeah. Me too. Right. I mean, yeah. Yeah. I think, and I, you know, real time simulations like that. Yeah. You know, bring them into the fancy IC design tools. It's like, you know, cadence isn't too fancy for that. I want to see the little dots in cadence too, because that's how I, you know. Bring it on.

Anthony Wall: Yeah. Yeah. Yeah. Well, that's, that's, I mean, that is really good to hear, even if it's not everybody doing it. If there's some motion in that way, that is definitely a silver lining of the shakeup that is the past two and a half years. Yeah. I think, I think that's. Many, many, many downsides of the past two and a half years. But yeah.

Chris Gammell: Yeah. No, that's, let's be positive. There were, there were some good, there were some good aspects, I suppose. Yeah.

Anthony Wall: Yeah. Well, Anthony, as you figure things out and decide where to go, we'd love to hear from you again and see what you're building next. Cause it sounds like you're going to be building some cool things in Silicon and in the world. Thanks for joining us here today.

Chris Gammell: Thanks very much for, for having me, Chris. It's been an absolute honor. We'll chat soon. Thank you. Bye.