Like every year, Heidi Barnes from Keysight Technologies stopped by our booth at DesignCon 2019 to discuss industry news. This year, the focus was on DDR5.
Here are the topics covered in the interview:
- 0:08 Do you have any comments on DDR-5?
- 2:06 Have you thought about the role circuit boards would play with DDR-5?
- 3:45 Does it make a big difference if the etch is more controlled?
- 4:22 What does happen when you shrink trace and space?
- 4:41 Simulation software are based on physics. What makes Keysight’s tools better?
- 6:43 How could machine learning play a role?
- 7:49 How often does a company need to invest in better simulation/measuring tools?
- 9:05 What design rules does the designer need to follow?
- 9:39 Do you want to say something about women in STEM?
Do you have any comments on DDR-5 and how an engineer can really digest and learn DDR-5 properly?
You know, welcome to the world of SerDes equalization. DDR five is finding out you can’t just brute force signals as they’ve done with DDR in the past, the single-ended, and you’re finding now with DDR five, it’s like we need to put some equalization in there and we’ll maybe we need some at the receiver end to a DFE and now you’re starting to talk SerDes. And the SerDes world learned a long time ago about trying to do channel simulations with convolving the bit pattern with the pulse response. They’re starting to try to figure out how to do that. Or actually, we have tools, key site. ADS has a channel, a bus simulator I call it, for DDR and it’s not a channel simulator specifically, it uses the same technology, but DDR has parallel clocking. It’s not clock data recovery.
And so, the traditional channels simulator doesn’t quite work. And so there are some extra tricks in there that you have to do with the parallel clocking and, but you use that convolution technique with the polts PRBS pattern, or not PRBS for DDR5, but you use at your bit pattern convolved with your pulse response and you get a lot of insights very quickly and you can see how that channel’s performing. So I think in terms of DDR5, it’s coming, it’s staying. Obviously, every time the, what is it, the cloud server farms get bigger, everybody’s into that business and more and more data going faster and faster.
Obviously, DDR has to keep updating and in improving the speed and density and everything. So with DDR-5, I think you’re just going to continue to see more and more of the SerDes techniques coming into designing and analyzing those DDR5 systems.
Have you thought about how the circuit board would play a different role or same role in DDR5 earlier?
Circuit boards, I think the fascinating thing to watch with printed circuit boards is the level of creativity going into the stack-ups and the control that we’re starting to see now of what materials, low profile copper, all your dielectric materials and drill patterns and drill. We’re getting away from finishing hole size. We can actually spell specify drill sizes so we know exactly what the size of that hole is for EM simulations. A lot of these better and more precise control of the fabrication process let’s the high-speed designers be more confident in getting repeatable performance every time.
And that’s, I think, part of what DDR is going to need because you have hundreds of lines and they all have to perform equally. So that becomes more of a control of quality across, say across a printed circuit board panel, or from panel to panel. So your product doesn’t change over time.
The other thing I think that’s kind of interesting is that, with DDR five, obviously, if you shorten everything down, you’re going higher speed to keep the same loss you just run shorter traces. But shorter traces also mean you can get more ringing between transmitters and receivers and you don’t have as much room to route things. So I think the challenge, especially for DDR five again, is looking at some of these advanced techniques that the SerDes layout people are doing. You really have to start looking at it very well controlled fabrication processes and design rules for the routing.
The etch on the PCB is controlled more or that makes a big difference if the etch is really controlled in the system.
Right, and in the RF Microwave world and high speed digital, we did that a long time ago. We try to stay with half ounce copper because now you’re at etch profile is better controlled. If you’re trying to etch down through two mils of copper on the top of your printed circuit board you’re going to get sort of, if it’s gold plated, you get a mushroom shape. If it’s silver plated, maybe you get the mesa type or, you know, it’s wider at the bottom of the trace than the top. And so a lot of those things, if you go internal to the board with strip line and half ounce copper, you get very good etch definition.
A typical trace and space is 3-mil or 75-micron etching space. And now the trace and space can go down to 25 microns.
If you shrink things, it goes faster. And I think as long as you can maintain the quality, shrinking things down basically does, without really changing much of the topology you can get up to higher frequencies. So that is sort of the fastest way to keep the performance maintained as you go up in data rate.
So I have a general question about the general simulation software out there. If simulation software are based on physics, what makes ADS better?
So HyperLynx actually is a little bit more of a hybrid there. They’re using some 2D or algorithmic models a lot of time to try to speed up that simulation. And I would say HyperLynx is great. Mentor is manufacturing printed circuit board layout tool integrated in with company production logistics and all of that. And you know, HyperLynx, you push a button and your layout designers can check things. But I think one of the challenges is engineers will often need a tool where they can really get in there and understand what part of a layout is controlling performance.
How do you get the innovation and the insight to improve things for your next design. How do you get that competitive edge with another company if you’re just pushing a button and not really understanding the engineering behind it. And so with ads, we actually let engineers look at the problem from different directions. It’s very easy to actually modify the layout in real time and hit simulate again for the engineer. And so, that’s our goal is to really integrate the ability to bring in a board, bring it into an EM tool, get out, export directly into a schematic analysis.
We have a new memory designer tool that is net driven, so you can bring in hundreds of net connections and make your transmitter and receiver connections with all of those bus lanes and then also run the EM model on that, bring it back into schematic, hook it up to your channel simulator. And then if you think you’ve found something in layout that you don’t like, you can go to the layout, change the copper around, and then do that whole process and an automated if you want to.
So the idea is to really enable the engineer to have the feedback he needs to gain insight into what’s really controlling the performance of the design.
Do you have any comments on how machine learning could change simulation?
So machine learning, you know, I think management at every company, every big company out there is trying to figure out how they can invest, how they’re going to make money, what it can do for them. Obviously, there’s a lot of people dreaming fast about how wonderful AI is going to just solve all of our problems. And I, you know, as an engineer, you know, where you’re trying to analyze hundreds of nets or you’re looking at all this data. You’re thinking, Oh, if I could just push a button and have an AI, just look at all that data and come back with the answer. It would be nice.
So I think in the future, definitely we’re every company out there, a lot of investment going into it. You know, you look at the chips coming out of Nvidia for AI, the automotive industry is definitely pushing that technology very fast, but the fact that the processing power is now fast enough to do the AI algorithms. And so you’ll see, I think, a huge bump in even our industry to try to figure out how can we start gathering all of our knowledge base and feed it into those engines to assist the engineers.
You know, we got lots of DDR5 memory and, you know, like the Nvidia chip. When I grew up, we used to have the encyclopedia Britannica and the local salesmen would come with these huge boxes of books and in one second now you can download the whole encyclopedia Britannica at the data, you know, at 100 gigabits per second. My God, it’s just crazy all the data that can be processed.
What is the threshold where a company realizes they need to invest in ADS or invest in better simulation?
You know, the obvious answer to me as well, you should, everyone should have this tool. If you’re an engineer, you should have, you know, an engineering toolbox. I think sometimes I compare it to the idea of if an engineer has a technician, they can do a lot of work together, and that was a model that we used to grow up with. And nowadays it’s, a lot of engineers don’t, we don’t get the luxury of a technician.
But if your simulation software is not cheap, CAD data, software, CAD software. But if you look at it from a point of view if you don’t have to hire another engineer, you’re augmenting an engineer’s toolbox. We have a great support network with our software. We link on also with our instrument partners and all of that support network. So when you invest in simulation software, you’re not just investing in a niche point tool to do something.
You should also consider the fact that that’s a resource. Get on that tech support hotline, ask questions if you get stuck and don’t be shy. That’s we’ve got a, we’re coining at key site care, and we’re looking at that future of how do we enable our customers to really make use of the instruments and software that we can provide them with.
If we were doing a thousand circuit board designs a year, what percentage of those designs use DR5 memories? Should we run all of them through ADS or when does it not make sense?
Well, obviously what you want to do is at first design, you want to understand the design rules. And so initially the first one’s gonna maybe take some time to figure out those design rules, but if you define exactly and understand what’s controlling the performance, then you can make very good decisions about what design rules the layout designer needs to follow. Then you know, you want to try to automate it more and then it becomes a much faster process to just run it through these tools for a quick check rather than having to actually analyze what’s controlling the performance.
Do you want to say anything about women in STEM?
As a female engineer in the world growing up, I never used to try to think of myself as being different. I think most of us, it’s human nature, you want to fit in. I’ve always liked tinkering and building and making things and I went to, Caltech. There were not a lot of women there at that time. And as someone I wanted to fit in, you know, I basically, I don’t know if I pick up some of the habits, but I feel quite comfortable in that engineering world and I don’t really see men, women, whatever. It’s, it’s icy engineers and I love the discussions that ensue from that. I was lucky enough, I did have a professor in college, Dr Rutledge. He’s RF microwave professor. He has a lot of famous students at different universities around the world.
But what was interesting to me is he actually, did make an extra effort to get women engineers into his research groups. And I did a senior thesis with him. One of his Grad students was Arthi Provocer. And she was there at DARPA also. So anyways, he was probably one of the ones that was very instrumental in helping in my career, you know, helping with women in engineering.
But I do, I think what I would like to say is I don’t really like to, to make those distinct… I would rather see women focus on the job and see us spend time going to technical conferences than conferences that are just for women. I think I do a lot more writing papers for a technical conference and putting on seminars and boot camps like I did yesterday to educate the industry rather than worrying too much about where people come from or what background I’m trying to move forward.