Issue link: https://iconnect007.uberflip.com/i/1441485
70 DESIGN007 MAGAZINE I JANUARY 2022 ered low-loss was Nelco 4000-13 SI, and they achieved it by using a low-loss glass. at worked just fine until the tsunami happened in Japan, and Nittobo, the one supplier of that glass, was inside the quarantine zone. Within a week none of us could build our products any- more because we were all single-sourced with that glass. I got involved with Isola to see if we could do something about it. It wasn't too long before we were doing 10 gigs, and it surprised me that we soon were doing 100 gigs. e current links, the undersea cables all around the world, are 400 gigabits per second, which is pretty aston- ishing. We are achieving that with eight links of 56 gigabits per second, which pushed every single thing about the laminates. e industry wants us to double that, and to do that we're using PAM4 (pulse-amplitude modulation), a four-level coating so that the clock is still 28 gigahertz. For us, the loss problem is still the same, but the noise margin, because the lev- els are smaller, is such that the error rate is not good enough right now. is is where I'm pushing material, now that we know the indus- try is going to want that. Between 1996 and 2021, we went from 10 megabits per second to 400 gigabits. at's a 40,000X increase in bandwidth in 25 years. To do that we have pushed everything about the laminate. Tarun is probably the most renowned at doing that because he's been the major player driving that whole thing. ere are two things that I worry about. One is skew, which is glass weave that causes the two sides of the differential pair to not arrive at the same time, and then the link won't work. e other is loss. Tarun, you can pick up the ball. Tarun Amla: I remember around 2002 where people said that five gigabits per second was going to be unattainable in copper. ere was a lot of work done around using optical chan- nels which did not get much traction. Around that time, the RoHS lead-free legislation was coming on board, and that just added another layer of complexity, because not only did you want these signals to go through at those high data rates, but you also wanted reliable per- formance. Boards now had to go through 6X reflow at 260°C for qualifications. In some cases, the qualification standards are about 10X reflow or even higher. Now I'm hearing about 20X reflow, which seems weird. Why would you want to do that to your board? We are at a crossroads where we have to look at the next generation. Sadly, finding an alter- native for copper is going to be difficult with the amount of work involved. A materials— dielectric—solution has to come through, or it will be like what happened with Moore's Law. We are at an impasse because we have reached the limits, especially with 112 gigabits per sec- ond. If you're trying to do it with PAM4, like Lee said, the margins are lower and you've got this inherent skew in that signaling, so that's where it stands. e whole industry is waiting, and whether it's 5G or 112 gigabits per second, the stage is set for something new to happen. Happy Holden: e people I worked with in Tai- wan making the flip chip substrates have been Tarun Amla