Rick Hartley – The State of Signal Integrity
We met with Rick Hartley during PCB West 2018 to discuss signal integrity, controlled impedance and other PCB designers’ pain points. What can you do to improve your designs? Listen to the master!
- When you’re designing, understand the impact on manufacturing.
- Fabricators need to be more proactive to let designers know what they can do and what’s available as a standard 4, 6, 8, 10, 12-layer stack-up.
- Insertion loss and attenuation are very important for designers, especially at high speeds.
- When it comes to power delivery, most of the energy is not at low frequencies, it’s at very high frequencies because it has to match the switching speed of the ICs output.
- The fields moving through the dielectric insert the current into the copper and the rougher the copper is, the harder it is to get even current flow through the copper.
0:10 What is the state of signal integrity today as seen by the design world?
I first started doing signal integrity work about 25 or 30 years ago in the late 80s. And at that time almost no one understood it and when I would do a class it would get a huge turnout. Today I don’t get quite the turnout that I used to because I think a lot more people understand today what it takes to do impedance control and why it’s critical. I think that’s a big part of it. People know why it’s critical and they know why they need to terminate transmission lines. They understand the board stack is a big part of it. And I think a lot of people know that. Where I think they’re falling down is they don’t necessarily understand the impact they have on manufacturing. They don’t always design to make things optimum from a manufacturing standpoint. And I have a feeling that you guys probably see a bit of that from time to time. Boards that could be more producible than they are. Or that could have better impedance control than they have based on the impedance they requested.
Learn about your PCB manufacturer’s capabilities
One of the things I encourage people to do in my classes is to learn about the boards that their fabricators make. In other words, if you’re going to get a 6, 8, or 10-layer board, find out what is the natural 6-layer that the fabricator wants to build. What are the dielectrics, what are the copper weights, find a way using that natural 6-layer that the fabricator wants to build, find a way to design around that board, design around those dielectrics, design around that copper weight. Even if it requires different line widths on different layers to hit some target impedance of say 50 or 60 ohms. Design around what the fabricator’s building and let them know in your fab drawing, “This is your 6-layer. This is your 8-layer. This is your 10-layer.” And if you do that they’re going to get a lot better throughput. We did that at L3 and we were getting high 90 percent throughput from our fabricators even on 12, 14, 16-layer boards that had very tight impedance control and worked well.
2:19 Would having standard stack-ups using conventional materials, like FR-4, I-speed, etc., be a benefit?
I think, yes, it would. And as I said, I encourage designers to find out what their fabricators are doing. Maybe the fabricators need to be more proactive to let designers know what they can do and what’s available as a standard 4, 6, 8, 10, 12-layer stack-up so the designers would know, “If I follow this methodology then I’m going to get better throughput, I’m going to get better impedance control because I’m going to be in their sweet spot.” And I think, yes, there would certainly be a place for that.
2:51 Are the signal attenuation or insertion loss kind of concepts important for designers?
Insertion loss and attenuation are very important for designers, especially at high speeds. Now, people doing stuff at lower clock frequencies, if they can slow down the rising and falling edges, aren’t as concerned, of course, you would expect that, but people operating in the gigabyte domain certainly have to pay attention, especially to attenuation and losses. In fact, I did a class at PCB West on that very subject. There are a lot of good materials, as you know, available for such a thing such as the I-speed material. I haven’t specifically used that exact material but I’ve used other materials with similar loss characteristics so I know the advantage, certainly, that that offers. A material like I-speed certainly can do that. There is one bit of good news for the design world on that vein, that being that the ICs that are being designed and manufactured today often have pre-emphasis and equalization built into the driver and receiver stages.
Focus on quality materials
The result is that high speed materials, probably you don’t want to hear this, are less needed today than they’ve ever been even though speeds are going up, because they can buy ICs that can handle the losses by altering the shape of the waveform either before it’s sent or at the receive end. And so, that’s a good thing actually for the design world. It doesn’t mean they won’t need high speed materials, because especially when they get into the 10s of gigabytes, now all of a sudden they’re going to need both. I’ve done design up in the 10s of gigabytes and you need to focus on quality materials and on equalization or pre-emphasis, you have to think about all of the above at that domain.
4:41 What is behind the idea of the signal that steps four times rather than once?
It gives four bytes with every transition. The idea of that is great. The idea is that by creating four different transitions that’s kind of like pre-emphasis because what you do, you change the amount of energy per unit time that you’re pulling out of the power bus and driving into the transmission line because you’re doing four time steps. And the idea is it allows you to reduce loss tangent concerns and skin effect loss concerns in high speed boards by doing that. And the other benefit of it is the EMI benefit. You have one clock signal for every four bytes, instead of double data rate, it’s quadruple data rate. The clock rates per given data rate can go down, which improves the EMI signature of the system. All of these things are benefits of that. The downside is you have a smaller noise margin per step, and so, you do have to be more careful with things like reflections, transmission line impedance control, termination sizes, all of that becomes more critical because you have a smaller noise margin.
And the length of the line can to some degree becomes restricted as well.
5:52 What is driving the miniaturizing of ICs? Why doesn’t the PCB industry keep up?
The way the IC industry is miniaturizing, the PCB industry has not kept pace. What’s driving that, of course, are smaller pitches of ICs, especially with BGAs. If you have a quad flat pack design, even if it has fine pitch, you can branch the leads out off of the pattern, big deal, but with BGAs, that is a big deal, it’s quite a problem because somehow I have to get all of those lines out from that very tight pitch part, and that’s a real challenge. When you get down below .5 millimeter pitch and under, .4, .3 millimeter pitch, it requires some pretty fine line technology. And I’m a firm believer that the fabricators who can step up to that world are going to have the future in their hands.
6:44 Do you feel there is a scope for 2 or 1-mil technology?
I feel there’s an absolute need for 1 and 2-mil technology. If you know Happy Holden you’re probably aware that when he worked for Hewlett Packard 20 years ago they were doing 1 and 2-mil line technology in some of their stuff, because they had to to be able to produce some of the products they were putting out. And so, yes, what’s happening now has gone from especially people like HP doing it to where the whole world is going to be moved in that direction whether they want to or not.
Physics don’t change, do they? Laws of physics, thank God, are constant and congress can’t affect them.
7:25 What perspective would you like to give designers who are involved in the automotive electronics, especially from a signal integrity perspective?
It’s an interesting question, what perspective do they need to understand? People in the automotive world still want to stay with very little layer count boards if they can. That’s key to them because low layer count means low cost. The first time I did a presentation for the automotive world in Detroit years ago, I was told by somebody from that industry that there’s an expression we have in the automotive world, “Who do I have to kill to save that penny?” And, of course, they don’t really mean it, obviously, but they are very serious about saving money. They will spend a lot of money on a design to take a few pennies out of the design. Layer count really matters and the IoT world, the Internet of Things world, is moving in that direction as well because they’re going to be 1, 2, and 4-layer boards, and that’s really where that world is headed.
Design for high quality signal integrity
People need to learn to design and still maintain high quality signal integrity in 1, 2, and 4-layer boards. And that is not an easy task. I’m starting to put together material on how to design IoT boards in low layer counts and still maintain that kind of quality and consistency that we now have in higher layer counts so that people in that domain can find ways to do it. There are ways to control impedance even in a 1-layer board. You have to do everything in a co-planer fashion, obviously. But you have to have a return path for every trace. Just as you would if it were a high speed board that was high layer count. And that’s the way the IoT and automotive and appliance world even are going to have to start thinking. Because as they get into reduced IC sizes they’re going to be thrust into the same exact problem with fine lines and low layer counts. And it’s going to affect everybody and they all need to think about this.
9:16 So fine lines will definitely have an impact on low layer counts.
You don’t necessarily need fine lines in low layer count, but when density gets huge and you still need low layer count, they’re going to have to think about that as well. As long as density is low they can go with standard line weights and widths and standard copper weights without any issues at all, but they still have to understand what it means to design the transmission line. They have to know that every signal coming out of these devices are fast. You mentioned that Dan Beaker was here, and I had a dinner with Dan three weeks ago in Detroit. He was telling me about an IC that a NXP Semiconductor recently redesigned. They went from like a 1 or 2 nanosecond rising edge down to tens of picosecond rise times at the output of this thing. And my question to him was, “Why did they do that?” And he said, “I asked the designers the same question, and the answer I got was ‘Well, that’s the output transistor that we chose to use.'”
Well, it wasn’t necessarily a good decision on their part, but the reality is the world’s going to be stuck with that like it or not, and the automotive world’s going to be stuck with that. And they’re going to be doing impedance control, they’re going to have to do field containment just like the high sped world is doing today, like it or not. Because regardless of their clock frequencies they’re going to have high speed signals because of edge rate.
10:36 Why do designers feel they need one ounce copper for ground and power planes?
It’s a matter of current flow. The problem is they’re missing the point. If you have very low frequency current flow, DC currents, for example, that are flowing… If you have DC energy that’s moving – and when I say DC I don’t mean DC voltages, I mean DC voltage and current because it has to be both or it’s not low frequency – people think that power bus delivery, you’re dealing with a constant 3.3 volts it’s low frequency. But no. The current is switching at hundreds of megahertz or gigahertz speeds. Just because the voltage is constant, the current is not.
When you are dealing with really low frequency current flow, when it is truly low frequency, they may need thicker planes for that reason. But what people lose sight of is that when it comes to power delivery, most of the energy is not at low frequencies, it’s at very high frequencies because it has to match the switching speed of the ICs output. And you have to provide all of the harmonics in that square wave from the power bus, from the clock to .5 divided by rise time, it has to be provide to the output station, the power bus. It is very high frequency. The result is that skin effect takes over at frequencies beyond a few hundred megahertz.
Watch for skin effect, current and copper
When they have rising edges that are sub-500 picoseconds, you’re now dealing with frequencies of one, two, and three gigahertz. The skin effect is going to be the dominating factor in terms of how much current the copper can handle. One ounce copper won’t do any better than half ounce copper in that domain. Once you go up beyond a certain frequency you don’t use the entire copper thickness. People are deluding themselves into believing that they need one and two ounce copper in planes. In fact they don’t. They really need to examine, “What is my current flow at high frequencies?”, then do the skin effect calculations and determine how thick the copper needs to be based on skin effect. And they’re going to find that very often even quarter ounce copper is good enough for planes. You don’t need one and two ounce copper in planes. And in fact, I ran into that when I was at AMD in the late 1990s, and the engineers there believed they needed those heavy planes. They didn’t, they didn’t. And I convinced them eventually that they didn’t need that.
13:00 What about the roughness of the copper and its impact on insertion loss, etc..?
Roughness of copper does have an impact, unfortunately, that is a true issue and that is a big deal. People tend to want to look at voltage and current when they analyze everything. Skin effect is about fields, it’s about the fields. Everything is about the fields, actually, but skin effect is one of those issues. The roughness of copper makes it harder for the fields to induce an even current into the trace and the plane because of that roughness, because the current actually comes from the energy in the fields. People think that the current and the voltage come from the driver. The fields come from the driver. And the fields generate the voltage and current in the transmission line or in the planes as they move through the dielectric. The fields moving through the dielectric insert the current into the copper. The rougher the copper is, the harder it is to get even current flow through the copper. And so, the losses become greater with rougher copper.
Instead of using a constant DC voltage on the drum, you use an AC voltage to put some on, take some off. And you end up with very low profile copper. I think people need to understand that they’re going to need to go to a very low profile. Or at least a moderately low profile copper in order to get the kind of insertion losses and skin effect losses that they expect to see in transmission lines.
14:39 In our trenched technology, we have found out that the roughness of the copper is not a concern anymore because you’re cutting the trenches by laser.
I see. You put down a sheet and you remove copper with a laser.
14:46 You remove the dielectric with the laser, then you plate, and you make a trench.
I get it. You’re plating the trench. I can’t wait to see this. That is really interesting. That is really interesting. The trench then is very smooth by definition, and so the plated copper is very smooth. It’s like the drum side of electro deposited copper. Yes, that makes perfect sense. That is brilliant. That is brilliant and that’s going to really make a difference in the behavior of high speed signals. People who are operating in the 10s of gigabytes are going to really appreciate that. The ones under 10 gig, 2, 3, 4 gig, who cares, you know? They think as soon as they go above a gig that the world’s going to fall apart. No, it’s actually above 10 gig where the world starts to fall apart.
2, 3 gigs, everybody does it these days, everybody’s doing that, even the automotive world, “Oh, shoot, that’s easy.”
15:35 What further advice would you give to designers today?
Understand the energy in the circuit is in the fields. You expect for your circuit boards to function as you want them to and to pass EMI testing? You must know where the fields are in the circuit. And you must design circuit boards, most importantly the stack-up. The stack-up must be designed correctly to make the fields go where you want them to go. This way, they won’t expand and won’t cause interference problems or EMI issues. And the best piece of advice I could give designers is focus on the fields. Ralph Morrison, who is now 92 years old, has recently released a book. It is called Fast Circuit Design and Energy Management. It’s a brand new book, I just bought literally last week myself.
I own all 13 of his previous books, and this is his 14th book. And it will tell designers what they really need to know about the management of energy in circuits. It will help control interference, EMI, and signal integrity issues. Very important publication.