Building Together: Oxide and Samtec
So I am really excited to have Jonathan Ghesh here and from Samtech. And, kinda before we get into this, I I wanna keep kind of a a somewhat long intro to kinda how we got to to Samtech and a little bit of of Oxide's approach. We talked about this in in other episodes, I know. I know we talked about this, when we had Kate on. We certainly talked about this when we had Doug on from Benchmark.
Speaker 1:But the you know, our kind of view about the way we build this stuff, is to us, makes a lot of sense, but it is a little bit idiosyncratic. So when we we know that we're building a a large machine, much larger than we can build alone. And we are not really looking for I mean, obviously, we need suppliers. But we we're really looking for folks that are gonna partner with us more deeply. And so one of the things that we do at Oxide is we really differentiate between partners and vendors.
Speaker 1:And a vendor for us is just selling kind of undifferentiated things. And, you know, the shipping is kind of the classic example of this. Like, we do not have a dedicated shipping partner. We just kind of ship on although, actually, maybe, like, what what is the one that, is it Pirate Bay? What what's in the Pirate Ship.
Speaker 1:Pirate Ship. Thank you. Yeah.
Speaker 2:Yeah. Pirate Bay or
Speaker 3:the other one? Pirate bay
Speaker 1:is the other one. Whoops. Yeah. Pirate bay. No.
Speaker 1:No. We don't do anything with pirate, but we do. And so I can maybe, I don't know. Nathaniel, do you use pirate triple hunt? Maybe that is.
Speaker 4:I mean, I I don't I I don't know, like, for dollars spent, how much we spend there, but certainly, a lot of us remote people use that a lot to ship because it's convenient, and you just print the postage and go.
Speaker 1:Oh, what's kinda heartbreaking if this is gonna be a really special relationship that we have with Pirate Ship because we're gonna have to find another example for something that is purely a vendor for us, but where we are are really not not differentiating. But for the for a lot of other things where we are really looking to go innovate with another company, we are really looking for for a partner. And we're really looking for someone who's gonna who's gonna understand what we're gonna go build it with us, and go take responsibility with us, with this thing. And, you know, I I don't know. I'd like to believe that that some of this comes from our own shared history at at Sun, where one of the, I think, the real strengths of Sun was the way that Sun viewed its its supplier relations and then also its customer relationships, and really valuing very deep relationships on both sides there.
Speaker 1:And we have lived, where those those relationships are not necessarily valued and where where the the suppliers are constantly being pitted against one another, and, and it's not good. So this is kind of our disposition for kinda how we build this thing. It's really looking for the right folks. And our intro to to Samtech, was Ariane, I I think it was when we were, really starting to to to figure out what the switch was gonna look like. This is in, like, the summer of 2020, fall of 2020.
Speaker 1:And, we this is the how hard can it be switch, which we are learning is, like, damn it. This thing is really hard. And somewhere along the line, Ariane, you were beginning to look at how we do act that SI from the actual switching silicon. We decide on Tufino at that point. But the actual signal integrity from the switching silicon to the the the actual QSFPs was actually gonna be a real challenge.
Speaker 1:Am I remembering that correctly? That that's kind of the where we kinda start on this journey?
Speaker 2:Yeah. So with the the the the thing I was trying to figure out was how do we how do we push the asek as far back in the chassis as we can so that we get it as close to the the sort of the back so that the the cables in the back can be as short as we could make them? While still having fans on the back of the chassis. So the board needs to be at a certain height in the chassis. And then you'll have these queues of p ports in the front.
Speaker 2:And I was like, how are we gonna resolve that? Because that is gonna be an enormous slab of PCB that sits in this chassis if you do that all PCB. And so, the the idea formed to or rather, I started looking at what what sort of solutions are available, and I stumbled upon the, the flyover cable system that Semtech has developed, which is a a set of connectors and then twin x cabling that goes from a QSFP interface or it can be, like, several connectors and interfaces to, to a connector using cabling. And then you can you can suddenly go, a, a little bit further because you, like, cabling has these types of TwinX cables have far less loss, much better performance than a PCB has. But more importantly, I could place, I could basically split the board into 2, and I could place 1 board at a particularly a particular height in the chassis that was accommodating for the fans.
Speaker 2:And then the other board could be exactly in the center of the chassis and then we could have QsFP ports on both sides of the board, so you can make them belly to belly as it's called. And then you would use these flyover cables to then fly back to the main board and land near the ASIC and then connect the 2. So that's where we started. So that was a solution that Semtech had developed. It was ready made, like, visible on their on their website.
Speaker 2:So, we reached out and started talking, with that as a starting point. And then, things, went from there.
Speaker 1:Okay. So and, Ari, have you heard of Samtech before kind of discovering this? I I certainly had not. Did my introduction
Speaker 2:to Semtech? I have. Because, because of the sort of quintessential SWD connector, that's a very prominent Semtech part. So I I'd heard of them before, and I'd seen some of their higher end cables systems out of sort of curiosity in the past, but never had a reason to design with it because, you know, I never had anything that would go that fast, until then.
Speaker 1:And so so you discovered that this kind of this flyover design, and I just remember you sending me, like, a pointer to this video, and I was like, really? It's like, oh, man. This is great. What a great idea. And, obviously, this is
Speaker 2:There was a picture on the website, and it turned out later that we learned that was Jignesh work. There's a picture on the website that splits a switch into 2 pieces with flyover cabling from the front to the with the ASIC more located in the back, for cooling purposes and we we wanted to do that both for cooling purposes and to keep the back the backside backplane shorter. And so it was like, oh, hey. This this this is clearly an idea that someone has, has tried because if it is on their website, it must work. That that may have been a leap of faith.
Speaker 2:I don't know. But, you know, it turns out it does work. But, yeah, that that's how we got started.
Speaker 1:And so then you reached out to Samtech. And the and, again, this is not a the company that I we're obviously talking to a lot of folks, and I just didn't really have any exposure to them. But then something, like, very different happened. Namely, I I don't know if if it was Nate or whoever we got a hold of at Samtech, but the Samtech told you, like, we're gonna ship a cable to you, like, today. And the I'd all I remember is that the next day, you're, like, they sent me this cable.
Speaker 1:Like, I've I've got it. Like, they sent it to my house. Like, I'm holding it. Yeah.
Speaker 2:I didn't know if it was quite the next day, but it was pretty fast. It was pretty fast.
Speaker 1:It was it well, can suffice it to say, like, we were having a hard time, like, getting, like, and everyone else we're talking with, we are it's, like, getting bogged down with NDAs and all this other crap. And all of a sudden, you actually have a company that's, like, oh, hey. Yeah. You wanna build this, like, crazy thing? Like, let's just send you some samples.
Speaker 1:And you're like, are you I I I did I, like, pass out? Is this actually happening? Are you an apparition? Are you a friendly ghost?
Speaker 4:They they are they are known for sudden samples. Right?
Speaker 1:They are. And so and, Jake Nash and Jonathan, so maybe this is a good time to to to get you involved here because, obviously, this is something that we learned is not an accident. And this is something that is a very kind of a deliberate approach that Samtech has to really support engineers. And, boy, does it pay off because you generate some real fans. I actually got a photo that I'm gonna drop in the chat.
Speaker 1:I know, Adam, you love having, annotating podcast with photos. I know it's your favorite thing. Best. Yeah. And so I've got a photo of Arian.
Speaker 1:This is, like, one of the great things with the pandemic is it made it kind of easy to take just, like, screenshots of meats. And this is, like, Arian showing me this cable and it just, like, this thing is, like, I've got it in my hands now, and, like, this is what it would look like. And it was really, really exciting and energizing. So, do you guys I mean, this is not an accident. Right?
Speaker 1:This is this is the you kinda take a very deliberate approach to kind of engineer getting engineers samples in hand.
Speaker 3:Yes. Absolutely. So So first of all, thank you for having, Jonathan and me on the podcast. Really appreciate that. So, yes, so Semtech's a very unique company.
Speaker 3:We we don't go after just the me too products. We want to go after unique products that solve a customer's problems. And in order to solve customer problem, first of all, you have to get customers of understand their problems and get them the parts that they need. So that's the philosophy. And, you know, you talked about sudden service or sudden samples.
Speaker 3:That's how the company got started. Where, you know, even what the low low speed and power connectors and all these these to send it just overnight, to customers, and we still do that as free samples. But over the years, as our products became, you know, on on the higher speed side or even, you know, very low small profile power connector, anything like that, The service has gone beyond just sending samples to also providing complimentary services and signal integrity for the full channel for the customers or doing cable management mock ups for the customers, for doing first order thermal analysis of fluid flow analysis for the customers. So we've taken that service, sudden service, to the next level now because we understand that in order for customers to use products that we make, which are not me too, they're unique kind of products in the industry, There's a certain learning curve that the customers will have to go through, and who better than us to help them because we have quite a bit of experience because of all the, we work with a huge variety of customers in different industries.
Speaker 1:Well, and and yeah. Before we before
Speaker 2:we dig deeper, I wanna stress that for anyone on on listening who thinks, woah, it's a cable. How difficult can this be? Let me absolutely stress that that a cable anytime you add a cable to a system, you now added a think of it as a third the complexity of a third printed circuit board or or or beyond that in terms of complexity. Cables are incredibly complex because cables cause all sorts of effects that you have to account for in terms of signal integrity, power integrity. You have to think about, your your emissions and because signals are gonna leave your board, and they're gonna go to some other system system or another board.
Speaker 2:And then you you have to take take care of that signal that that rides on this cable. I've now been privy to the design of 2 cable systems, and it is absolutely some of the hardest things you can do. So when Jignesh talks about providing services like signal integrity and channel modeling and even, some fluid dynamics and things about around heat like temperature and heating and all that. That's absolutely necessary. And it is very much appreciated when when this is done by someone who actually understands what they're doing because it is it can make or break your system.
Speaker 3:Yeah. One Yeah. Let's say, just one way I like to tell and, Arlene, you talked about earlier. But, you know, in in traditional systems, you know, the the conventional way of making designs was on the printed circuit board. And what the way I think is is you're designing flat land.
Speaker 3:A PCB is just 2 d space, maybe a mezzanine board, but basically you're designing flat land. What cables allow what and as you mentioned, it allows you to design in 3 d space. And all the boards don't have to be at the same level. You can make it modular, you know, your, and you can even rotate the boards 90 degrees if your architecture allows it. So they they don't have even the same even, plane or direction.
Speaker 3:And and for that to happen and to take the advantage of that, traditional conventional architectures is we were the electrical engineer. Electrical designer was the chief architect, and all the mechanical guys and the guys just had to adapt after that. I think as the data rates go up, the electrical architect, the mechanical, the thermal, the manufacturing architect, all need to to sit together and they need to compromise together, you know, for a system to be successful, just like yours.
Speaker 1:Yeah. That's very interesting and interesting.
Speaker 2:Definitely true. And we
Speaker 3:we have been we
Speaker 2:it has been a balancing act between making it work, just getting a cable system to work with with with low enough loss so that you have enough signal left over on the other side to use it and recover it and actually drive the link. But then how to fit that in your chassis and how that runs and how it is manufactured because and both I'm sure we'll touch it on this in this conversation or and and we touched on it with Doug. Manufacturing a cable system of of complexity like this, is its own whole different thing. Like, how like, you could you can have all these mock ups, and I can have all these pin outs and spreadsheets. And what these what these spreadsheets don't really convey is that because you're only taking, like, a a small slice of a pinout out of that backplane.
Speaker 2:But that's that's thousands of signals, that all have that all land on these connectors. So you're you're looking at tens of connectors, hundreds of connectors sometimes, and then, you know, it it it it's it's quite the the the juggling act to it's a high wire act to make it all fit, ultimately, and then make it manufacturable and repeatable. And
Speaker 1:Well, that's it. And I think it's it is really helpful. And then it's just the thing in terms of Semtech's general approach to really be walking with you through all the kind of these phases, and kind of exploring these various tradeoffs, and not being kind of left to, merely do it on your own, which has been, I mean, it's been huge for us. And, and Janesh, I thought it was very funny because when you came out to Oxide and saw the switch, and saw what we done with the flyover cables, you're kinda, I just, you know, you were telling us, you're just like, love this design. Right?
Speaker 1:Well, it's honestly, like, we're just doing what you, this is your guidance basically, Semtech's guidance. And we kind of assume that every
Speaker 5:You just build what was on the website.
Speaker 3:We we It
Speaker 1:was on the website. And we just, like, assume that, like, isn't everybody doing it this way? And you're like, no. No. No one else is doing it this way.
Speaker 1:We tell everybody to do it this way. This is the right way to do it. But no. No. No.
Speaker 1:This is not a, so, yeah, I guess we we we just felt it was a it was a great idea. It's obviously a great idea. Everyone should do it this way.
Speaker 5:We preach the the concept and the philosophy all the time, obviously. And, you know, some people get it, and some people say, oh, okay. Yeah. That's nice. But we're gonna go back to doing it the way we've been doing it.
Speaker 5:But there's you know, obviously, you guys saw the benefits. You can put plate you can put things on your boards. You can split boards, different levels. You don't have to like, you guys have a full cable backplane. So that is to not have a a a a backplane PCB, I mean, that's a ton of design effort.
Speaker 5:Totally. Board spends. I mean, that's just incredible, savings. Just that one component in of itself. Not to mention everything else that we bought already talked about.
Speaker 5:But, anyways, we we we talked to we talked to people about this concept all the time. And we've had, like, we you know, we've had some customers that are building smaller chassis or smaller cabinets, you know, not full size cabinets like you guys did. So when we saw your cabinet, full size all the way up and down, full rack size, and it was, you know, completely cabled, it was, you know, not not just the cable backplane, but the all the sleds were cabled, and it's just like, oh my gosh. They did it.
Speaker 1:Well, and and truthfully, we had an unfair advantage. And, Adam, I remember you and I talked about this very early on in Oxide's life that we knew that we were gonna have an unfair advantage just by taking a clean sheet of paper. And, I don't know, I can't remember if we talked about this explicitly or not. But when Adam and I started a startup inside of Sun, in 2008, we what just by or 2006 when we started it. We just by by taking a clean sheet of paper, we were actually able to deploy Flash more or less earlier than any other enterprise storage unit, just because we were walking up to it with a clean sheet of paper at the right time, effectively.
Speaker 1:And I don't know. I can't remember if we talked about it or not explicitly, but I remember thinking, like, there's gonna be some equivalent to that, where we are just gonna have the benefit of having the clean sheet of paper at the right moment, and everybody knows that, like, hey. There's this way to do it over there that is the better way to do it, but no one has a clean sheet of paper right now.
Speaker 6:No. That's exactly right. And this is the perfect analogy for, you know, Flash in 2008 was just changing, and we were at the right place at the right time. And this similar example. And, actually, we bumped into a bunch of other ones.
Speaker 6:We're just not having to pull along this long chain of legacy. So let us do a lot less, actually.
Speaker 1:Totally. And being able to do I mean, Arne, you've been able to do exactly what you guys described of of actually being able to innovate with respect to to the the switch. The switch is actually not one board. It's 2. You've got this QSFP IO board that is then actually separated by an air gap effectively, with these flyover cables, and the distance actually to the QSFP board doesn't hugely matter.
Speaker 1:Right? We I mean, we could because once you're on the the kind of the cable superhighway, it it it's the loss is all gonna be at the Yeah. It's
Speaker 2:it's all pretty pretty minimal in in like, it doesn't matter whether you're gonna well, yeah. Like, over those distances, it doesn't matter. Like, it it starts to add up when you go, you know, 3, 4 feet. But in this case, we're going, I don't know, to an e some inches. So it's not so bad.
Speaker 2:But so what's interesting though is that we started this because I saw that picture, and I was like, oh, so Jignesh pitched it because of I think the main draw was like like a thermal,
Speaker 1:thermal. Yeah. Yeah.
Speaker 2:That's perfect. Properties. Because in a traditional switch, you have these QSFP interfaces that sit in the front, and then you need to put your ASIC, like, right there next to them because you wanna keep those PCB traces as short as you can in order to to stay within the loss parameters. And this is a real thing. A 100 gig, a 28 gig NRZ, it's it's, you know, you don't have much.
Speaker 2:But then when you go to 56 gig PEM 4 for, you know, 200 gig links and 400 gig links and beyond, The PCB is just eating so much of your budget that it it it it it you have to be super close. But the problem is that all of these QSFP interfaces, all these these, transceivers all have a a pretty significant heat load. They're you know, they started, like, 3, 4 watts, and they some of them run up to, like, 6, 7, 8 watts. And but you have to design for up to 10, so you have to accommodate potential, like, long range modules that can go all the way to, like, 9 or 10 watts. And so if you have this 300, 400, 500 watt ASIC that sits right behind a load of QSFP interfaces that is also, you know, 100, 200, 300 watts of power, You have a very large amount of heat, like, in a in a super small space.
Speaker 2:So the the the the thing they gave us was like, oh, you can put this the ASIC much further in the back so it's closer to your fan, so it's easier to, evacuate the hot air that your ASIC produces. And I was, well, that's that's a really nice feature, but I was more interested in keeping the the signal path to the backplane short. But the thing that fell out of this was that because we we now cut the board into 2 pieces, we actually defined an interface between these 2 boards. And these 2 boards were actually designed sequentially. They weren't even designed so much in tandem.
Speaker 2:I designed the interface alongside sort of the the first board, and I had a pretty reasonable idea what I wanted to do between these two boards. So there's flyover cabling for, the high speed interfaces back to the ASIC, and then there's a there's a bunch of low speed signaling between the two boards to manage the the power on that on that front IO board. And there's a little couple other components like LED drivers, and there's a there's another 5 for copper ports on the front. And so there's there's a little there's some management in there, like, pieces that are that are there. But what it let us do is it basically let us design focus on the main board and and get that interface done.
Speaker 2:And then once the main board was already on fabbed out, so that was already in production, the PCB was on its way, and assembly was on its way. That's when I I turned around, and then we built a QSFP board that would attach to it. And the q and we we already started bring up of the main board. We started that in January, and I don't think we've had to had the QSFP board in hand until February or early March. But that was fine because we had plenty of work to do on that main board.
Speaker 2:So we could sequence these things into, in in a pace that sort of made sense for us without having to design all of it upfront. Because if we had if we would have had to design all the qsoup pieces Oh my god.
Speaker 5:Yeah, in
Speaker 2:that main board, we would have had to push the main board would have taken another month or so to get done because the QCP board was actually, like, quite, like, far more complex than I thought. Well, because Totally. Because
Speaker 1:before? Like, how hard can that be? You're like, oh, Jesus Christ.
Speaker 3:Yeah. But it's still
Speaker 2:that's still 2 and a half 1000 components on a PCB, and it's it's it's a really thick PCB because It's a what? Because you have to, like, mount these things belly to belly. But there's because you want all these interfaces, all these ports, you want power control over them, and you want fault protection. So if if your your QSFP module shorts, you don't want that to take out, you know, 4 other modules associate like, next to it. So because of because of a power, like, irregularity or whatever.
Speaker 2:So each of these ports has a power, has a little eFuse that we can we can use, a, measure power, but also to control power. We can we can turn it on, off, and it and it makes sure that if a module faults, it immediately shuts the module down without rippling through all the other modules in the system. So there's that, which which adds all the stuff. There's all these ESD protection diodes that we need to put on there for for various like, for for because it's these are exposed interfaces on the front, so you need all this protection. So, yeah, that would have added easily like, it took us a month or so, month and a half to get that design all done, like, all buttoned up before it was all done and said and done.
Speaker 2:And so
Speaker 1:And the flyover cables were just huge to allow that to kinda be Yeah.
Speaker 2:Because you can you can cut that off. You can cut the interface between these two boards, and then you can you can sort of delay the, you can you we we were able to push the QSFP board out. But now the the kicker is that we designed the QSFP. So the flyover the pin out and the flyover cables, we designed such that we can actually respin this QSFP board in a different configuration. So today, we easily support, 32 ports on the front, 32, 200 gig ports, but we can reconfigure this board with actually very minimal effort into 16 400 gig ports.
Speaker 2:And then, we can use the same connectors on the main board to then fly these 400 gig ports to the main board and make those work without having to redesign the main board at all. That same interface would work. It it's we basically we basically would do half the half the QSFP board and but then with the QSFP DD interfaces so that we can accommodate QSFP DD modules. Or if there's another standard that comes out at some point, we can we can do a different form factor altogether. I have this idea of maybe at some point, we'll build a QSFP board that has a GPS receiver on it so that you can do very precise timing for
Speaker 1:it. Right, right, right, yeah.
Speaker 2:So not all not all switches in your network need to have that, but you wanna have a couple that might have that, you know, a couple per data center. And so we can have a special flavor of the QSFP board that has a GPS module built in so that you can connect an antenna to it and you can run GPS and you can have an accurate time source in your network. And so all these different variations of that board we can build without having to redo the main board. So it be it becomes a compartmentalized problem and will allow us to build several different variations of a switch chassis, with relatively minimal and and and relatively low risk, engineering effort.
Speaker 1:And so we were hugely sold on the flyover cables. That was going really well. And then we were initially, we and I, you know, I I hopefully I'm in a safe space where I can confess this. We were not actually considering Symtech. I mean, Samtech for us was kinda like, alright.
Speaker 1:We're gonna use it for the flyover cables, but we're gonna look for, another vendor for the cable backplane. And but as we Because we
Speaker 2:were already in progress with that other vendor.
Speaker 1:We were already in progress with that other vendor who
Speaker 2:is And to be clear, we were trying to get them from being a vendor to being a partner, but they could get them there. Yeah. We couldn't get them there.
Speaker 3:But, I heard I heard that then Jonathan happened.
Speaker 1:Is that Jonathan happened?
Speaker 2:Yes. I mean, so the the the problem was that we we were we were I was trying to get drawings for these backlink cables using exomex connectors. And, and we were promised drawings, and the, you know, the date passed and there were no drawings. And then we would try it again, and we would they would be there next week and next and the week after, to the point where, we and so out of the out of some like, some some reluctance for me to get started to because I had already figured out, like, oh, you are a second source of these these exomax connectors, so we might be able to make this work. Now there's a big difference between there's a difference between what is advertised in the in the catalog because the catalog of
Speaker 3:of of
Speaker 2:connector and cable manufacturers is vast, but what is actually tooled up is a different story. So there's always gonna be you may not have the exact same thing, but let's let's like, you know, fortunately, you did. But I was initially a little bit reluctant to to immediately jump ship because I did I I that was a big gamble, I felt, and and, I didn't wanna pit necessarily 2 companies against each other because also that would have been, splitting my attention, and I did not have much time to go. And because I was spending an awful amount of time I was spending a lot of time with the other company trying to get this to work. And so I was I was worried that I might have to spend, you know, similar amounts of time to get this other to get it to work with you.
Speaker 2:And I was not really in favor of that. So I was actually I was actually the one who had to be, like, pushed and pulled kicking and screaming a little bit because I didn't wanna I I didn't wanna do that immediately. But then, yes, we started talk then Jonathan happened. And we actually you managed to lap even though you you started probably 6 or 8 weeks behind, managed to get the drawings in hand before the other partner vendor managed to do that. And I suppose we have to tell them, like, we you just you're just late.
Speaker 2:Yeah. I'm sorry, but you're you're just late. What you did?
Speaker 5:I I remember some of those initial conversations, and they were they were very interesting to me because I was actually surprised that the other company was was kinda being slow or slower than what you were hoping they would be. We actually have a good we actually have a good relationship with them, so I'm not gonna talk bad about them. But Samtech does try to do obviously, sudden service is something that we try to provide. And I could tell that you guys were frustrated. And, you know, it was from the very first call, you know, you kinda made it clear that you you're you were kinda behind schedule from where you wanted to be.
Speaker 5:And so I just I just wanted to take that as an opportunity to run with it and and get stuff to you as fast as I could. So
Speaker 2:Well, the problem was I I I like, I again, I don't I also don't wanna speak ill of them, but the the the the problem was that there were just too many people involved in the process. And what I what has been incredibly helpful working with you and Semtech in general is that there's a very small number of people that we work with, and you bring in additional individuals when needed. And that that really helps in terms of getting these things done because
Speaker 1:it means
Speaker 2:that there's just not a whole lot of overhead in in in in the way that things need to be communicated.
Speaker 1:Well, you know, my mom has an adage that I'm I'm sure is is the time memorial that nobody cares about your money like you do. I think she was telling me this shortly after she just lost a bunch of money to a financial analyst who had who, and, you know, I've I've always taken that to heart. And the the part of the problem that we were having or the frustration is is, like, we just were not getting someone on the other side to treat this like their system. It was our system. And they were kind of it was just they they felt very reluctant to actually and, Jonathan, when you kinda jumped in, you treated it like a system we were building together.
Speaker 1:And that was really, really important. Just like that level of of attention to detail and empathy and let's go kinda figure this stuff out. And that's when you started out, yeah, it was, like, 8 weeks behind or more on something that was really time critical for us. And, you know, within no time had basically passed and was in much better shape, and we weren't in in a position to really be like I mean, it was it was a no brainer ultimately. And a no brainer because the just the degree to which you cared about our problems being shared with the 2 of us, that, like, we are gonna take responsibility for building this system together.
Speaker 1:And I just can't tell you how much how meaningful that is. I mean, that that especially when you're when you're us doing things that are a little bit, little bit little bit odd. You know, the, as our our Nice.
Speaker 5:Thank you for the kind words. Let me know the address where I can send you a gift certificate for saying such good things.
Speaker 1:It the the yeah. Exactly. That's right. If you could strike out the categories. The no.
Speaker 1:No. No. But the the in action oh, listen. The thing that is is so interesting to me is when we look around the industry and as we were kinda adding folks to the company that had been in you know, that with a lot of experience, like, they came in with very strong relationships because you have helped them. Samtech, the larger Samtech has helped them in past gigs.
Speaker 1:I mean, RFK, Robert Keith really wanted to be here today, and he is, as you know, is a is a huge Samtech fan. The only reason that he's actually at a graduation, like right now, and even then he was like, so I I made this worst possible time. Sorry, RFK. You're gonna catch the recording. Even then, he was like, let's see.
Speaker 1:When's the graduation start? And this is like a there is a a nonzero chance that RFK is at a is literally they are they've got, like, pomp and circumstance, and he's on his earbuds right now trying to unmute himself to contribute to this conversation. Because that and that is not, like, a normal feeling for engineers to have about but it's it it really shows the partnership. And I also love, by the way, that we, you have now like set the bar for our expectation and we had, well, let's just say a DRAM vendor that was trying to impress Oxide by coming by bringing gifts. The problem is the gifts they brought were alcohol.
Speaker 1:We're like, we don't want any of this. Our case is like, I want samples. That's it. Like, take I don't know. I'm not even freaking booze.
Speaker 1:Like, I think yeah. I don't know. Go take that home or something.
Speaker 4:I would I wanted samples. Build.
Speaker 2:Give me samples. Give me give me technical data. I mean, we need to get this done. Then we can Yeah.
Speaker 5:There you go. Contraption. So what I what I wanted to say though is we've really enjoyed working with you guys. You guys are a very good partner for us as well. And, you know, for all the folks listening, I think that it's important to understand that Oxide does a good job and how they do
Speaker 3:a good
Speaker 5:job for us. You know, they're buying cables from us, but they didn't just say, hey. We want this cable. Here's a signal map. Give us this length.
Speaker 5:Give it to us. Go. Go. Go. You know?
Speaker 5:It was an open discussion. There was a lot of talk about, hey. What's the optimized signal map for how you make how do you make these cables in your production facility? You know, is there an optimized way that you would lay out the cables from a signal map standpoint? We we because we're early enough in our design process that we can accommodate that.
Speaker 5:And then we talked about cable links and sleeving, how things go together. And Jodie, could you talk about And just really listen to, you know, what we had to say about how we were making the cables and accommodated that on the design side in your system. And that's really you know, that helps facilitate a lot of learning what your vendors like this.
Speaker 1:Yeah. Could could you elaborate specifically on good bends versus bad bends? Because I think this is a very concrete example of what you're talking about about, like, not all bends in cables are the same.
Speaker 5:Yeah. Certainly. We I know we've talked about a little bit before. You guys have talked to him a little bit about before with with Doug specifically when he came on from Benchmark. But the the TwinX cable itself is, you know, you can picture it.
Speaker 5:It's 2 center 2 solid center conductors, and that's surrounded by a dielectric. And so in essence, if you're gonna bend that that twinax cable, a good bend would be if you can picture, both of those center conductors bending at the same time with the same bend radius. A bad bend would be, you know, one of those being on the inside and then the other one coming over top of it with a larger bend radius. Because that in that case, the bend is putting, is pulling and pushing on those center conductors relative to each other because one of those radius is is smaller than the other.
Speaker 1:And so we call the gamma question?
Speaker 5:We call that we call that a it's both.
Speaker 1:Okay. Yeah. What
Speaker 5:happen is it will it will push and pull on the center conductors a little bit, which will put stress on the cable endpoints, whatever it's connected to. And then from a signal integrity standpoint, yes. It's also potentially bad because now your signal paths may have different lengths and especially relative to each other. And with the differential pairs, you know you know, it's all about optimizing and keeping those channel links as close to being exactly the same as you can possibly get.
Speaker 2:Yeah. You're gonna get a little bit of a PNN skew, which means that you're gonna have some potential inner symbol interference because at those speeds and the lengths, that stuff sort of adds up.
Speaker 5:Yep. I mean, in the grand scheme of things, it's, you know, it's very, very small. And because of the construction and how we make our cable, it's, I'd say, it's minimized. But, you know, just like Ariane said, I mean, you're you're getting to speeds where every little bitty thing is starting to count. So
Speaker 1:Yeah. Every it's
Speaker 5:doing it's doing the little things that are setting you up for success as opposed to finding the needle in the haystack when something goes wrong.
Speaker 1:Totally. And you I mean, one of the things I feel I've learned on this is, like, you just wanna be nowhere near your margin. Mean, you wanna make sure you've given yourself lots and lots of margins. Kinda like the Visa bill. Any little loss in signal is gonna add up, and all of a sudden it adds up to a total loss signal, and you've got no no way to kinda get that back.
Speaker 1:It'd be hard to get it back. So, Jonathan, do you wanna elaborate? And we I know we talked a bit of when we had Doug to from Benchmark, talking about cabling back when we talked about this, episode that we had, at Benchmark where we were we were getting some dropout rates in our cabling. And could you talk about that from the Samtech side? Because I was actually even, you know, being, loving our relationship and being super positive about Samtech, I was actually taken aback by just how quickly you all jumped in on that.
Speaker 1:So could you describe a little bit that problem from from Samtech's perspective and kinda your approach on that?
Speaker 5:So so I think this is this the you're asking about the cable that where you were having signal integrity issues in our end and we were
Speaker 2:Yeah. In in Oakland? We found these cables that had we had some damage in these cables where, like you said, because of stresses, the solder joints had cracked because of the way that we'd assembled the rack. And we we we didn't know yet how how to treat these cables properly to make sure that you you don't get that. And so, unfortunately, I wasn't able to make it, but you joined Doug and and Robert in at Benchmark Minnesota to to think this through and figure out how to do this better.
Speaker 5:Yeah. I was I was out in the Bay Area. Oh, man. Was that was that Design Con? Yeah.
Speaker 5:That was a Design Con time. Yep. Yeah. And I was I was able to go out and visit some customers, and and, outside was high on my list of of people that I wanted to come see. And it just so happened that, you know, I think leading up until that point, there was a cup a cable or 2 that was was, you know, showing signs of signal integrity issues.
Speaker 5:And so when I got there, you know, I started talking about, you know, just like we were talking about, this this bending cables in the right direction, making sure other things like you have generous bend radii, you know, not putting wire ties on until the very end of the of the process so that you're not trying to connect a cable that's wire tied somewhere and you're pulling, you know, you're pulling the end of it while the while it's wire tied tight somewhere else, these types of things. And so I remember I had been traveling for quite a bit, and I was sitting there talking with the team and and in in Oakland, and everybody like, yeah. We gotta get we gotta get the benchmark, and we need to, you know, just take a look at a whole unit and look at how it's wired up. And I'm like, yeah. I said, we I agree a 100%.
Speaker 5:You know, we gotta go, and we need to get in front of a rack and just take it apart. And we started talk you know, spitballing different ideas. And, I remember Steve coming over, and he's like, yeah. Let's make it happen. He goes, let's do it next week.
Speaker 5:And I'm like, man. Next week, my wife's gonna kill me. He says, alright. No. How about 2 weeks?
Speaker 5:We'll do it in 2 weeks. I'm like, my wife's still gonna kill me, but okay. I'm like, if I'm not there, I'll get somebody to be there. And Steve just looks at me and goes, no. We really want you there.
Speaker 5:I was like, okay. I'll be there. And, of course, I think he's got the screws. I think it got I think it got pushed out a week, from there. So I think it ended up being 3 or 4 weeks later, but and so the timing worked out just just fine.
Speaker 5:Didn't have any personal issues. But once we got into the benchmark oh, sorry. Go ahead.
Speaker 1:Yeah. Well, Jonathan, it is worth just a quick aside because you're based out in New Albany. Right? I I mean, Samtech is in
Speaker 5:New Albany.
Speaker 1:Yeah. In in Indiana. Right. And our manufacturing facilities is in we're at with Benchmark in the big ten engineering. We got a lot of big ten alums, and, and, one thing I've definitely appreciated is that the big ten engineering, we got a lot of big ten alums, at at Oxide, and there is a lot of terrific capability in, you know, when people talk about kind of onshoring manufacturing.
Speaker 1:I mean, we we did we that that wasn't a kind of a a policy position for us. That was much more of, us being pragmatic and finding real partners, and those partners just happen to be onshore. But in this case, it was a really big one. I mean, if, you know, I I think, if you had to go on a, you know, a 12 hour trans specific flight, that probably would have been a lot less appealing to go to a manufacturing facility.
Speaker 2:So No. But that was exactly the reason why we are doing onshore manufacturing because we just didn't wanna spend all this time traveling back and forth between the US West Coast, the US East Coast, and and Asia because it's
Speaker 1:just And that you and
Speaker 2:how many times consuming to to debug systems and debug your processes when you have nothing because we have nothing yet. So everything is being built from scratch, from our product, to our processes, to our tools, to our manufacturing capability, all of that had to be stood up. And so when COVID hit, we did deliberately say we want to do this in the US because it'll be just so much easy. It'll be so much harder to travel to Taiwan or to China every time we need to do something. And again, Robert or RfK spent so much time in China that he was like, I don't want to do this again.
Speaker 2:Like, it's it's and which I'm really glad we did because, I'm
Speaker 1:really glad we did. Oh my god. Can you imagine?
Speaker 2:I I I actually think that a little bit of a side thing that in our case, onshore like, manufacturing in the US is actually cheaper than doing it overseas because of because not losing all this time and not having to spend all this back and forth and traveling and and and all the like, all of it basically to to make this happen, that has saved us time and money. And therefore, I think it is a more efficient solution of whether that's still true in the long run when you start shipping, you know, many more units, we'll see. But at least for now, it is very well it that that works for us very well.
Speaker 3:According, it is so good for to hear that. I think, if it's a very labor heavy anything that's extremely labor heavy, then, of course, you know, you want to be in the lowest, cost place as far as manual labor is concerned. But when you design things in in a modular fashion where things can be reused, where the asset where the whole assembly process is optimized from the where when you start from scratch with a piece of paper, as Brian mentioned. I think you things can be designed in a way where, you could easily manufacture this very competitively in the US, especially in the Midwest.
Speaker 1:Well, totally. And you think of, Nathaniel, I mean, how much time have you spent? I I mean, obviously, okay, too much time in Rochester. But, like, that I
Speaker 2:mean, it's The source architecture is too
Speaker 1:much time.
Speaker 4:No. We've had, I mean, we've had an engineering presence there almost continuous for the last 6 weeks, and, like, you just probably couldn't pull that off overseas. And, you know, and, like, with a few of us who live in the Midwest, it's, you know, a few hours over there. So in an emergency, you know, it's you know, know, you can get over there for a couple of days. And, I mean, if you're gonna spend a couple of days traveling, you know, to get somewhere else, it's not really an option to go for just a couple of days.
Speaker 1:Well, it is so, yes, and that is true
Speaker 2:at larger firms. Like, for example, I was working on Oculus products before. We had engineering presence in at the factory in China, but it meant that those were individuals we hired in China. We're located in China, and still a significant portion of our team was traveling for every bill to be present there. It was a huge, like, amount of extra, like, the overhead is the wrong word.
Speaker 2:It's like it it it puts a real strain on the team. And for a team this size, it's just not feasible.
Speaker 1:So, yeah. Anyway, Jonathan, a long way of saying we were really appreciative in that moment of having that it was easy for you to to really get to the manufacturing facility and look to see, okay, let's sit down, you know, with production engineers and with our mechanical engineer, with with Doug, with and and really kind of figure out, like, what what's going on here to figure out, how is this damage occurring?
Speaker 5:Yeah. Absolutely. Yeah. And I, you know, I didn't finish, but we took the cables back that, were were, suspect and basically tore them down, looked at them. And, yeah, the solder joint, you could tell, had been cracked, which is evidence of pushing and pulling on those center conductors and just overbending or putting too much stress on them somehow.
Speaker 5:And it I mean, there's no real telltale sign, but other than just, oh, you know, too much handling, if you will, Yeah. That's what we had kinda come to. And so that was, you know Well, your late this is the thought it was so important to to get up there and and, go through a system with you guys.
Speaker 1:Well, in in your latency from actually taking those damaged cables and actually getting them I think you getting them imaged, like, x rayed them or not. You could getting them imaged and and and inspecting them and just I mean, how I mean, just for so many things that I've had, Adam, how many things that we had in our career that gets sent back to a manufacturer only to be told that there's no fault found.
Speaker 4:I mean
Speaker 6:Yeah. Exactly.
Speaker 1:And it gets so frustrating to hear over and over again. There's no fault found. So the and it just feels like did you actually, like, try to find a fault in this thing? Or are you the and the so for you all to not just not only, like, actually, yes, find find damage, but find it quickly. And I think this is like, it just goes to you the disposition.
Speaker 1:Like, your disposition was not how can we exonerate Samtech here? Which honestly feels like the disposition often in the industry is like, you've sent me this failing component. My job is to prove to you that this was not my fault. And it's like, that's really not your job. Job.
Speaker 1:It's like, we we, us together, we have something that has failed, and we need to understand why that has failed so that we together and it's like just a lot of people don't have that disposition, and you very much did in terms of, like, getting understanding where's this damage coming from, and then where could it be coming from? It's like, no. This is not like Samtech's fault, but it's like this thing is busted. How did it get there, and how can we help brainstorm different ways to handle this thing, manufacture it such that we don't break it? It's like that is so refreshing.
Speaker 1:It's a very refreshing disposition.
Speaker 3:Yeah. I think luckily, the physics we use is the same as the physics you use and everybody else uses. You know?
Speaker 1:You know, that's not always the case though. I feel like I'm alternating in a different physical reality than a lot of these vendors, where they're like, so, yeah, thankfully, the same physics for sure.
Speaker 4:And I would say I was there on-site when Jonathan was up, and it was just really awesome to see, Jonathan and Doug and Eric and RFK. Every like, everybody roll up their sleeves, and they're all, you know, elbows deep in the rack, pulling cables out and rebundling things and, you know, talking through how we might change the the way our design mechanicals are to facilitate, you know, these nicer bends. And so it was just it really felt like a whole team kind of all standing there working together even though, you know, it's like 3 different companies.
Speaker 1:Three different companies. Yeah. And that is and, Jonathan, I'm, I mean, I I'm not sure if this is par for the course for you or not, but I think it's that's really special from our perspective when you when you're able to really have that kind of collaboration across a bunch of different folks who are outside of their own self interest and really looking at, like, what is the how do we kinda build this thing and get it working together?
Speaker 5:Yeah. I think it comes back to what you, you know, what you said a couple times is we're we're building this thing together. And, you know, it's we we certainly have some some customers that are just, just, you know, give us the spec. Give us this. You know?
Speaker 5:Just throw it over the wall to us. But when when we have customers like Oxide that take the time, wanna fully understand, want want and ask additional questions, looking for that support, we're gonna give it to you. And you guys do the same thing with other vendors and, know, even with your customers, and we do it with our vendors. So I think we're all of the same same mindset.
Speaker 1:A 100%. And you're exactly right. That I mean, they we have very much that model with our own customers. Because it's very important that, like, you know, like your your customers, like, oxide, we're a technical customer, and oxide's customers are are technical. And, you know, we are building something together.
Speaker 1:And in our case, we are helping them to deliver a service to their customers who are they are often in the walls of a company, sometimes out outside the walls of the company. But, ultimately, we are helping them deliver something. And we wanna be sure that if if that if that's not working for them, that we have all the right people in the room together to figure out how do we fix this for you? How do we write that for you? And Adam just said at the top, I feel like that's something we really inherited from Sun.
Speaker 1:I do think that that is something that Sun got really right is we always felt empowered to do the right thing by our customers. And and we I never felt like we had to justify doing the right thing by a customer.
Speaker 6:That's right. And and that's how we spent a bunch of our early careers with customers making things right or or fixing things or or making things right even when, just as in this case, you know, irrespective of where the the blame, as it were, lay. But how do we make them successful with our product, whether it's our bug or their bug?
Speaker 1:And, you know, honestly, like, we're so used to this that it was kind of foreign to me that any company wouldn't like, why wouldn't you not do this? Right? Obviously, this is the only way to do business as far as I'm concerned. That you of course, you're always invested in your customer success. And it was a little bit jarring.
Speaker 1:I got so I don't know. I don't get the same same disposition, but I think I think I got, like, so used to that. That when I first kind of encountered companies in the wild that didn't have that disposition, it was, like, pretty shocking. And, I remember in particular, we I had a a comp I was demonstrating DTrace, which Jonathan Gignesh, this is a technology that Adam and I developed together to allow you to dynamically understand what a software system is doing, to basically answer the question, like, why is the software system not performing? And, generally, this is very positively received, because who wouldn't wanna know what their software system is doing?
Speaker 1:But this particular customer received it very very poorly, and in particular, they didn't want their customers to be able to run this on their software to understand how bad their software was. So the and I'm not gonna review in this meeting or this meeting did not go well. I was not really braced for the a customer, in particular, the customer was like, you have to disable DTrace on on my software. Like, I will not support my software on your platform until it cannot be instrument with DTrace, which was not something that I I'm
Speaker 6:sure that was a good opening salvo. I'm sure you toned that down afterwards.
Speaker 1:Yeah. I don't think I've ever had to be escorted out of a meeting by the sales folks who I it just did not go well. It went really, really poorly, in part because we had spent the first half of the meeting where the trying to understand why their software is performing so badly on us, and it's because they were shipping debug a debug binary with debug symbols. And it was when I ended but they told me, like, sorry, like, we can't rebuild it. We can't optimize it.
Speaker 1:And it was when they told me that they wouldn't certify Etrace, that I volunteered, that I was gonna write a tool to allow their customers to extract their source code from the binary, because it had all the stabs information. Like, the source code was all in there.
Speaker 6:Yeah. That that sounds like the right time to escort you from the room.
Speaker 2:Yeah. It clearly did learned a little bit. But so to to do bring that back though That's right. Alright. That's right.
Speaker 2:I think that a lot of a lot of, companies don't necessarily do this on purpose. It's not that they don't want to help a customer. I think that for some reason, in a lot of places, the people don't have the ability to do it because the either the information or the knowledge is spread so wide and so, but not in a positive way that it's it's difficult to actually find the people who can and bring them together to solve a problem, let alone if you have to do this with 3 2 or 3 companies. And then there's a bunch of politics involved with all that, which makes it usually, like, adds another whole another dimension to it. And so, therefore, the only response that a company can then have is, like, this is not our problem.
Speaker 2:This can't be our problem because, of course, we designed it properly. Because if it was not like, if there was if there is a problem, then they don't have the ability to figure out what is wrong and how to fix it, which is a pretty bad place to be in, obviously. But I don't I'm willing for a lot of places to not have it be on purpose. I think this is just sort of like what happens. This comes back a little bit to the other cable manufacturer that we were trying to work with.
Speaker 2:It wasn't that they didn't want us to succeed in that sense. It's just that they were in their way of having too many people involved to even get a set of drawings out and to give us feedback on a pinout and on how this mechanically was going to fit. It wasn't they want it was not that they didn't want it because they clearly wanted our business, I think, or at least I hope so. But they just couldn't figure it out between all of them.
Speaker 1:Yeah. They told
Speaker 3:Yeah. And and then what's unique about Samtech in that case is, you know, it says Jonathan is actually the product manager for all backplate products. Right? And and I'm a technologist and for multiple of the high speed products. We don't only give internal direction on what the next generation architecture should look like, but we're also involved with customer issues on existing because if you understand the problems that you have on existing products, and we can't deal with all customers, but, you know, at least leading customer, then we can design next generation better.
Speaker 3:And, usually, in the other companies, you don't deal it's not the same person doing both, and that's the
Speaker 1:And Yeah. That's the critical piece. Important point, Indranesh.
Speaker 2:You need to be you need to be subjected to your own decisions and the consequences of your own decisions. Only then will you learn how to improve, like, how to how to make your how to make your product better. Yes.
Speaker 1:Well, and this is something that I believe very very strongly in that you should take responsibility for your ideas all the way from that initial ideation, all the way to actually running in production. And it is really important to do that because you then I mean, it cannot underscore enough what you're saying in terms of, like, wait. You should not view it kind of pejoratively. Those things that are running in production, those systems have a lot to teach you about what about your next generation designs. But you have to go listen to you have to, like and the other thing that I'd be, you know, the the the thing that I I you hate to hear, but but happens way too frequently is folks being told that you're the only one seeing this problem.
Speaker 1:This is another thing that, like, that and it happens way too frequently in the industry that that, you know, you all never do because it's like, it doesn't matter to somebody if they're the only one seeing the problem. It doesn't even matter if it's true. It's almost always false. But, like, it doesn't even matter if it's true. Like, never tell somebody you're the only one seeing this problem.
Speaker 1:Because what you're effectively saying is that, like, the rest of the customer base has somehow outvoted you. Important. Yeah. More important. Yeah.
Speaker 1:And it's like the the it and, you know, there are other if if what you're trying to phrase is like, wow. This is really unusual. Like, say that. You know? It's like or I you know, I need to understand, like, what is in this environment?
Speaker 1:What makes it I need to understand why we're seeing this here. That's fine. But, like, when you're telling that you're the only one seeing this problem, it's like, oh, god. So frustrating. And it just shows, Jignesh, how many how few companies take your approach of actually having folks that are coming both from the design side going all the way into the customer base and actually understanding this thing.
Speaker 5:Yep.
Speaker 1:And I feel like, Adam, this is the the the this kind of is dovetailing it into their our our cattle bladder episode, which I can't even remember. It's a better title of. I just remember that. I was able to use tackle bladder in casual conversations. That's all I remember.
Speaker 1:That would we're talking about I think this is the I know this. This is UNIX episode. But we talked about customer support and why and how important it is to actually be to to really walk with your customer with their problem.
Speaker 5:I think it was interesting. Like, some of my very early conversations with Oxide, I had a couple, like, 1 on 1 calls with Ariane after I'd sent him some samples, and I think they were just, like, mechanical samples that like, even down to the component level. And I remember Tom Wenarian, like I don't know. I'm on East Coast. He's on West Coast, so I think it was, like, Friday at 5 o'clock.
Speaker 5:And and we're he's asking me questions about all these just minute level detailed questions about the components. Not just not like like the overall cable assembly. It's it's like down to the the copper that we use. And I was just blown away because I I think at one point, I told him, and this is still true, I don't think I've had any other customer ask me the level of detailed questions that Ariane has asked me about our cables.
Speaker 1:Really? Oh, interesting. Wow.
Speaker 5:I mean, I don't get involved at some of the Jignesh is, like, more of a technologist if you heard him say that in. So he's he gets involved in, like, some of the super far out conversations. I'm I'm just a lonely old product manager. But still, I talk to I talk to customers all the time, and I still have not had that level of detail. And I guess where I was going with that is that it's just the level of you know, the the level of detail that goes into, and I'm sure you do this, and I know you do this with other vendors too.
Speaker 5:And that's helping out that partner relationship, kinda closing back up to that loop. And and that's just so important.
Speaker 2:Well, it has a very specific reason why I why I ask those questions because so I took all your samples and took them apart. I cut things open. I I I pried out those little, like, little metal tabs that go inside of connectors and then took all the wafers out and looked at the wafers and scraped some of the wafers off and desotted some pieces. Because one of the things I really don't want to have happen is us designing a cable backplane. And at some point, a customer coming back and be like, hey.
Speaker 2:This we don't trust this cable backplane because we see failures, or we've had some of these conversations already with some of our customers or prospective customers. And the first thing they they hear when they hear we're doing a cable backplane, they have in mind these connectors with straight pins, and they think about bent pins and how one one unit goes from one slot into the next. And, you know, as a virus goes through your rack and you bend out you bend everything, your cables, your connectors, your chassis. Like, all of it is at some point broken. And so I really want to make sure that I understand how this thing mechanic works mechanically, electrically so that we make these decisions appropriately and that I can ultimately stand behind this design.
Speaker 2:Because I don't I just don't wanna be called out of bed at 3 in the morning because these failed systems have failed. I've done enough pager duty at this point. And so I
Speaker 5:don't want you to either because you'll probably call me.
Speaker 3:No. But it it has
Speaker 2:been really helpful because because, for example, there there there is there was a particular engineer or a particular person at a particular prospective customer we're talking with. And this person kept on pushing back on his cable backplane. And I made a short presentation for specifically for him explaining how these how how the XMX connectors work and how you have these counter like, these counter opposing fingers that then that that insert and that basically get pushed away from each other, and that's how you get tension between them. And that it is very difficult to, if used appropriately, to get these connectors to be bent or these fingers to be bent and cause problems in other connectors that like, where they are mated. You may not have a connection when these things if you bend these fingers far enough, then sure, your connection will work, but it won't damage other connectors that you're inserting these into.
Speaker 2:And so being able to explain that and being able to I took these things apart. I've I've I've looked at this. This is how it works, and then we worked with Doug to make sure that our mechanical solution has enough float in all the directions so that, you know, when these things made, that they made appropriately, not not too fast, not and that that we they made far enough, all that all that good stuff. Because I wanted I just wanted to make sure that we we really dotted our i's and crossed our t's here because we're we're we're betting to farm on this. Yep.
Speaker 2:There's no going back at this point. We can't. And you were talking you were talking about a back line out of made of PCB material. We just can't. There's not enough signal and take it like, there's not enough, insertion loss budget Yeah.
Speaker 1:We just can't do this
Speaker 2:out of a PCB. It is it is gonna be this or bust.
Speaker 3:Yeah. I think now you're hooked. Right? As in once you design the cable, you're never going back.
Speaker 2:Well, actually, now I want the next thing. I want I want a fiber optic link because then I because I I can't push I can't push copper much further than what we're going today.
Speaker 3:I I will disagree with you on that, and we can talk about some other time. But Yeah. Copper, every time people say copper has every at every data rate, people say that the end of copper now is fine. Optics and copper Let me
Speaker 2:let me put the asterisk here. Unless we can move the connector closer or even onto the package of the ASIC itself. If you can do that, then suddenly we have 7, 8 d b extra extra loss budget. Yeah. Then we can push copper again further, higher later.
Speaker 3:And and and Samtec already has one that we've we we got our first ending samples now where the density of that connector is better than even any silicon photonics engine that is available to the market. So just to give you a perspective, it's it is 64 differential pairs. So think of it as 32 t x and 32 r
Speaker 5:Yep.
Speaker 3:In half inch by half inch footprint.
Speaker 1:So can you elaborate a little bit, Tignesh? Because I think this is something it's a question that we do get from time to time. People like, well, isn't aren't optics the right way to do this? Right? Actually, like, not it's certainly not right now.
Speaker 1:It's it it's copper. And I think that that you're I saying that, like, people really underestimate copper and that, you know, know, everyone says that kinda copper's gonna disappear, but actually, I could could you elaborate on that a little bit? Like, why is copper still, like, the the the right way to build these kinds of systems?
Speaker 3:Okay. I think it's good to look at not just from the signal integrity point of view, but just overall system and your customer point of view and your and our our customer and your customer point of view. If you have a switch, what the current front panel, you know, like QSFPs and DDs and OSFPs allow is for you to ship a switch without optics in it, for example, for people to without opt and then the customers can put whatever optics they wanted. So customers will always prefer to have front panel optics if possible. That is 1.
Speaker 3:Yep. The the second one is what I am talking about earlier with the current, things where you don't want to keep your temperature sensitive optics very close to a high, powered heater aka the chip. So if you can separate the 2, the life of the and the reliability of the optics is significantly higher. There is, an erroneous law which says, you know, as the temperature increases by 10 degrees Celsius, the laser's reliability decreases by 2 x. So you want to keep you wanna keep the optics as low temperature.
Speaker 3:So there is a there there are there are reasons why you want to stay up keep optics away from the chip. Right? And and so and copper is just a dumb passive, copper cable to dumb passive. They they they can withstand the same temperature if they're higher than what the silicon can. And so if you can do that, I think people will always will try to if if there's some technology that allows you to do that, you still have transcribe optics, they will still go for that.
Speaker 2:So so there's that there's that. Then the other part the big part for me is power consumption. Because the reason we don't do optics is because your chip inherently has copper signals coming out of it because that's like, there's there's there's wire there there's bumps under your under your like, if you go, like, all like, the dye itself has pads. There's bumps on those pads. You go onto a package or whatever.
Speaker 2:That is copper by definition. The so if you go to an optical engine that is off chip or even on package but still off dye, now you you're spending you're spending power to go from one interface through a little bit of substrate or whatever that is into an optical engine, and then optical. That optical conversion process obviously costs power, and then you go out. So that that is ex that that usually that because that will lose against any if you have a passive copper solution, then, yes, that will always be more power efficient because that you don't do that conversion. Now that might change a little bit once you start talking about these integrated photonics that go on
Speaker 5:the chiplets on the package.
Speaker 1:I've got Adam, I have silicon photonics on my bingo card. I assume you do as well. I I'm not sure.
Speaker 6:Do you have a bingo? For me, for sure.
Speaker 1:Okay. They're very exactly.
Speaker 2:Well, this yeah. This is where you would integrate the photonics into an actual chip piece that then goes into your onto the same package as your ASIC. And now some of these these these equations change a little bit because you can use the lowers the lower power interfaces between the 2 chiplets. So you're you're talking now stuff like what you see AMD do with their chiplet designs for their CPUs and their GPUs. And, I I I don't know whether that changes enough because you could now potentially shave off the SerDes and not not do this conversion twice.
Speaker 2:And now you you you could do it in sort of one step. And so I don't know. So at some point, there isn't there is a crossover point where that starts to make sense. But then we we would have to start thinking about how to blind mate optical links, which is, you know, I think doable. Just it's all, sweat and tears to get it done and money.
Speaker 2:Right.
Speaker 1:With the with the but the PowerPoint is also a really good one because it's, like and I I know this is a concern that Eric had when we were just regarding compliance that we are not gonna have thermal monitoring on those QSFP ports, and we're gonna burn the thing down. And you look at, like, the potential power consumption of those things is really high. So this is actually I mean, you know, get as as you said already in at the top, we got, you know, potentially, like, 300 watts out there in in ports. So, it it's, it's a it's a good point about that lasers are actually, as it turns out, pretty power hungry. And there's there's a real power argument to be made once you're you're already in copper.
Speaker 1:So staying in copper, if you possibly can, is gonna have a bunch of wins, in terms of thermally and mechanically and so on.
Speaker 3:Yep. And, Arun, to your point, right, if you can convert directly from the package itself, that saves a lot of loss. And the other is, can the can the copper cable to an x itself be made lower loss? And the answer is yes. Our next generation cable, you know, Jonathan talked about that it has a dielectric around the 2 copper conductors.
Speaker 3:It's a solid dielectric right now.
Speaker 2:Mhmm.
Speaker 3:And and so what we are working what we'll we'll be calling it to an ax air, but, basically, what we're doing is we are putting air voids in that solid dielectric so that the net dielectric is significantly lower. So the loss in these cables would be around 20% lower than the current cables.
Speaker 1:So as
Speaker 3:you as you go to the next data rate of 224 gig, that's where the next generation cables will start becoming more
Speaker 2:So so we're basically trying to recreate the the ideal physical model that we're all being taught in school or that you're taught in school and then. Yes. Or like suspend 2 copper conductors in absolute, like, space in in Right. Like, low low temperature low low temperature vacuum, speed of light. That's
Speaker 5:what you model.
Speaker 3:Yes. Yeah.
Speaker 1:Nice. We still have that frictionless surface.
Speaker 5:Yes.
Speaker 3:Yeah. And so we have, like, our first engineering samples of those type of cables already available.
Speaker 2:And
Speaker 1:Just very on brand. It's like they add and, actually, if everyone looks under their chair right now, you'll see see the sample has already left a sample for you. You get a sample, and you get a sample.
Speaker 3:And this will be all 3
Speaker 5:of them.
Speaker 1:The the generic side, that is honestly amazing that there's another 20%. I mean, that's a huge difference in, and you know, kind of a, you, again, Arin, you made this point earlier, but like, you know, cabling, like how complicated can it be? It's like actually really K but really complicated and really, really, really important
Speaker 3:Yeah.
Speaker 1:And really important because it's like, boy, you think that that 20% loss that has a 20% improvement in loss has got ramifications for power. It's got ramifications for SI. It's got all these ramifications up and down the system.
Speaker 2:Because the the the 20% is is is combined with because every generation of these systems, the receivers get better, meaning the receivers turn more and more into radios, meaning they're bigger because you need more silicon to more, and you're actually using analog circuitry a lot to make these receivers more more sensitive, meaning you're spending more and more of your silicon area to to do this. So this this is another reason why chiplets has something become become such an interesting thing is because these analog that you want in these SerDes have very different, silicon properties than your digital logic, which is, like, super small and you want the latest node. But for your analog circuitry, that doesn't really work that all that well. So there's a different trade off that you need to make. So at some point, cutting these in different pieces and making them fabbing this on different different nodes in order to basically get the best out of each of these interfaces, it will become important.
Speaker 2:But as these interfaces become as these receivers become better, you know, another if you if you can eke out another 20% on the inter on on the receiver or 25%, then all these things stack up. And that that that does mean that you can suddenly do these higher data rates at these longer lengths of cable that we want to run them at Because our our our channel our total channel length is is is the the cable is about, about 6 feet. Is that 6, 7? 6 feet. That's the longest channel that we have, and we play some games with making sure that the longest cables are paired with the shortest traces on the print circuit board so that we borrow basically some margin out of the printed circuit board and put it back in the cable.
Speaker 2:And likewise, so, all the these are all games that we play in order to make that work. And so but, yes, I'm a little bit hesitant to start thinking about 112 gig link, because that that's Yeah. Well, because we just have to do more modeling. That's that's what it Right. What it comes down to.
Speaker 2:And these blind mate interfaces become more important. Like, the consistency of blind mating these will become that that that's where the where the this goes, because you need to be able to mate them consistently every time the same way within the specification of the connector. So I don't know. Fun challenges.
Speaker 1:So Nathaniel is saying that Silicon Phonics has been 1 year away from mainstream for quite a few years. I would like to thank you for acknowledging that it's been 18 months away, which is the you know, we when someone is saying something's a year away, that's a little too tangibly close. It's it's when it's they're talking, like, 18 to 24 months. Like, alright. That's just far enough away to not be verifiable.
Speaker 1:And Yeah.
Speaker 4:I joke, but, you know, I've been doing this for 16 years, and I feel like silicon photonics has been 1 year away from mainstream for most of that.
Speaker 1:A long time ago.
Speaker 4:Who you talk to. And so, you know, I mean, there's there's interesting stuff there. And, I mean, certainly, the like, things like triplets and stuff give that more legs because you do have that problem where it's like your new ASICs wanna be on, you know, super high end tiny little transistors, and then your photonics maybe want some different technology there. So anyway.
Speaker 2:But I do think you think Photonics are gonna be this one thing in my career. The the the the the all these different manufacturers are gonna hold this in front of me. They're gonna dangle this carrot every time they're designing something, and I'm gonna get hopeful every time. And, like, I don't and I I'm I'm already,
Speaker 3:well, I'll
Speaker 2:I'm just gonna go out and say it. Intel has already stomped on my heart twice now, and so it's it's painful.
Speaker 1:Intel, are you listening?
Speaker 4:Right? Should should we tell him that we have a pre meeting mailing list that we tell all of the vendors to dangle silicon photonics out in front of him? Yeah.
Speaker 1:Nathaniel, I told you, you're not worrying about that. It's a it is important that that we we know that listen, Ariane's easy. He's a but I I'd actually you know, we Ariane, we kinda tease you about it, but I love the you know, like, constantly, like, looking forward and kinda getting creative is is so important. It just feels like silicon photonics is destined to be a heartbreaker for a long time.
Speaker 5:And, We we weren't gonna tell you this, but we have a silicon photonic sample. We'll send it to you, Armin.
Speaker 2:I don't know. I know you do. I mean, I I I I you you even gave me the, the 20 gig NRZ one that that the what is it? The Firefly cable, and I'm I'm sure that tech has developed further. But what ultimately, the the the important bit for that, it's like, honestly, for silicon photonics to really work and to go somewhere, it needs to go on package.
Speaker 2:It needs to be paired with the dye, like, on package, not as a septic separate optical engine that you still need to get through through the PCB because that just that by itself, you're going through the the the the BGA package, the balls into your PCB, back into your optical engine. You've already lost, like, the game. Like, at this point, like, why bother? So which means that the only way this will happen is if we can figure out how to get this on package, which means that we need to get, you know, the the the switch basically We need to
Speaker 1:make yeah. Exactly. Semtech. We need to get Semtech to make switching silicon.
Speaker 2:Or we need to do our own switching silicon and then invite Semtech to go put the
Speaker 1:Oh, there we go. Engine in there. Yeah. So And maybe we did drink that booze after all.
Speaker 3:You know,
Speaker 1:I thought we gave it back.
Speaker 3:All, I think oxide is the would be a is the perfect partner for Symtech. I'll I'll I'll just I think, the the the DNA of both companies are very similar. So that I think that we are a great partnership. And, also, for the 224 gig, I'm not even talking 112 gig and that, Firefly connected and talk about that course is co packaged. We are partnered with quite a few, of the major chipmakers, and a customer of theirs because it requires all 3, a system vendor, chip vendor, and connector vendor to work together.
Speaker 3:Yeah. And and that is happening right now, but, you know, there are it's it's a it's it's a struggle also at the same time because I'm getting 3 companies to talk to each other, especially if they're very large companies. Some of them, it's different.
Speaker 2:Well, not just very large companies. They all have 3 different roles. They've played 3 different roles in the industry, and they've never had to cooperate in this manner until now because the ASICs the ASIC development has always happened a little bit in isolation. And then, you know, you just get the data sheet and you get the device and you make it work on your on your board. And then similarly, the the the, you know, the system integrator builds out of pieces that they that they that they find and and and sometimes commission.
Speaker 2:But in order to now you need to yeah. You're right. You need to think about the ASIC design designers and and and the we're out of the package designers. That needs to be planned. And that but that means that you're now subject to these, like, 5 or 6 or even 10 year cycles that these a six are on.
Speaker 2:And now you need to intercept that, and your system integrator needs to intercept that. It's it's this is gonna be really interesting how how and and I wonder if this is actually going to be caused to for some of this stuff to slow down because some of the like, some of these partnerships, I don't know if they will be able to get to get done to be quick.
Speaker 3:So I think that's where the customer a a company like yours, I actually has an advantage because you have not constrained yourself by the form factor of a 1 RU switch, actually. Right? You call it a whole new right? You have it as a for you and a half, like, half width and so you're defining your own form factor. Yeah.
Speaker 3:Oxide unit. You're defining your own form factor so that the overall architecture makes sense. And I think that the approach that needs to be taken overall. And and so you already are are one step ahead of most of the system vendors in that case.
Speaker 1:Jigesh, I gotta say this just warms my heart because we are so used to people being like and I feel like you're almost like when you're talking about oxide doing its own form factor, this is where other folks would be, like, looking for euphemism to explain, like, we're basically nuts. Or it's like, as you know, oxide does not feel constrained. It's like, yo. Are you saying that we're weird? We've done but we definitely have done things our own way, and we think for very good reason.
Speaker 1:And I you know, it's been terrific that Samtech has has always, just dug into the problems and understood why we've done it this way. It's been such a great, as I said earlier, a with a really perfect partnership. It's a model partnership, for us. We definitely point other people to the the Samtech oxide partnership as as really one in which we're innovating together. And, for those of you out there, if you haven't, if if you're hearing about Samtech for the first time, as you may well be, it's a fun website to disappear on, and there's a lot of great content out there.
Speaker 1:And, Jagadish and Jonathan, thank you very much for joining us. Really, really appreciate it. And, Arian, thank you for, thank you for finding them. Thank you for, you know, taking the approach that you did. Yeah.
Speaker 1:And it's it's been it's been a lot of fun, and, it's been a lot of fun to go actually, go build this thing together.
Speaker 2:I honestly don't know how we would have built it otherwise.
Speaker 1:We would exactly. We would not we would have been we would not have built it. So
Speaker 5:Yeah. Yeah.
Speaker 3:Thank you very much. Man. I'm based in Santa Clara, and we have a lovely demo room that shows how all these different architectures can happen. Anyone wants to just stop by, we can do a quick lunch and learn for anyone. Just reach out to Jonathan or
Speaker 1:me. There you go. Awesome. Well, and I yeah. I'd encourage folks to take you up on the offer because it's definitely our most alright.
Speaker 1:Well, hey. We're not gonna be we're gonna be off next week. Adam, I'm actually giving a talk on social audio next week, which I'm
Speaker 6:yeah. I I so You should have us all there live commenting
Speaker 1:on the chat. We should. And we actually I it's good. It's almost unfortunately, my I think my position is slightly earlier in the day. This is go to go to Chicago.
Speaker 1:But, really looking forward to be talking about what we've done here with Oxide and Friends. So, if there are particular things you want me to hit on, definitely, hit me up in the DMs on that. So it's been this has been so much fun to do, and in part because of, just like this discussion today. It's been a lot of fun. So thank you, Jonathan, Jing Nash, and Jonathan, and, and Ariane, and Nathaniel.
Speaker 1:And, we'll see you in 2 weeks.
Creators and Guests
