ATSs Qpedia Thermal eMagazine is pleased to be a Media Sponsor of the Electronics Thermal Week, co-sponsored by MEPTEC and SEMI-THERM. As part of that series, we are developing some posts around the speakers, issues and suppliers involved. This is our kick off article, an interview with David Copeland, the Keynote Speaker for The Heat is On MEPTEC Conference being held during Thermal Week.
David Copeland is a Thermal Engineer working in the Packaging Technology department of Oracle’s Microelectronics Group, developing packaging and cooling technology for UltraSPARC processors and the systems which use them. Areas of development include thermal interfaces, heat spreading materials, single-phase and phase-change liquid cooling, and data center cooling. He received his BS from Massachusetts Institute of Technology, MS from Stanford University and DrEng from Tokyo Institute of Technology, all in Mechanical Engineering.
David’s Keynote Topic is: Energy Reduction and Performance Maximization Through Improved Cooling. We had a chance to catch up with David on the phone, to ask him to let our readers know about why he chose his topic and what he hopes his listeners will take away from his presentation.
ATS: David, why did you choose this topic?
DAVID: Today, thermal engineers are given a semiconductor and told to maintain it at a certain temperature, 85 C or whatever, and I think this is true for those in other areas of thermal management as well, such as LED and Photovoltaic. Here’s a device or component, now create a thermal management solution that keeps the devices junction temperature at the specified point.
But now, it looks like temperature is being a variable rather than a constant. If you look at the International Technology Roadmap for Semiconductors, they have the high performance microprocessor junction temperature dropping from todays 95 C to 70 C by the year 2024. It’s only an average of 2 degrees per year decrease but taken as a whole, it’s a large drop that thermal professionals have to design to.
The other trend we are see is regarding power density. Power density may more than double over that same time frame of 2011 to 2024. So, as thermal engineering professionals, we are getting a bit constrained in terms of the higher power density and the lower head room.Now, if your environment is, say, a 45 degree C environment, that drops your head room down from 50 degrees to 25 degrees, so that’s exactly half the temperature differential. So the job has become somewhere on the order of 4 to 5 times more difficult.
The magnitude of the thermal resistance of the path from junction to ambient will have to go down by a factor of 4 to 5 over the next 15 years, which is not out of line with what has happened in the past 15 years, its just that our profession doesn’t currently have a lot of easily implementable solutions to get there.
The other thing we need to consider is microprocessor leakage current. The relative sensitivity of temperature to leakage current is becoming very strong. For example, in some process nodes you might have had to drop the temperature by 40 to 50 degrees in order to successfully cut the leakage power in half. And now that number is getting to be more like 20 to 25 degrees. So that means that the ability to make a given percentage cut in power consumption is slightly easier now with improved thermal management because you have to go down a lower temperature differential to get the same energy savings. Of course the concentrated Photovoltaic and LED lighting also have their own performance vs. temperature curves as well in terms of efficiency. Everything runs better when it’s cooler. But, how much cooler can you afford to make your device or system and still come out ahead; that’s the question.
ATS: David, what are some the strategies to get us there?
David: The thermal management migration path general trend is to lower the total resistance of the path from junction to cooler. And that is being done in two ways.
One way is improving the conductivity of the material being used and ultimately removing some of these materials. For example, one strategy implemented a few years ago was the introduction of solder thermal interface material. And that rapidly became wide spread. It made its way into Intel and AMD main stream processors even. Its a very good example of much less conductive material that was typically filled epoxy being replaced by a higher conductivity material.
And we’ve got some improvements in the heat sink as well. We don’t often see aluminum heat sinks very often now, except in the lower power devices. And so we’ve migrated from aluminum to copper and now we see embedded heat pipes and vapor chambers pretty much standard even in the x86 space, and certainly in the high performance market.
The next step could easily be lidless packaging which is already being done by a few people. It’s common in the notebook market where it is as much for space as it is for performance.
We also need to mention GPUs as well, graphics processor units. GPUs on cards also fall into the category of processors. The level of sophistication of thermal management of GPUs has actually surpassed that of microprocessors. And if anything they are on a much steeper growth rate in terms of power than microprocessors are. GPU manufacturers have much tighter control over thermal solutions, with the heat sink pre-attached and in many cases an integral blower, so they don’t have to worry about contract assembly houses not properly implementing a thermal management solution the way the computer space tends to. So the whole assembly process and the heat sink and the whole cooling solution is under the control of the graphics card maker. And that allows them to be a bit more adventurous and creative in terms of a solution that might otherwise require multiple parties to buy into a strategy.
ATS: It would seem that the more of the system an OEM controls, the more they can remove unnecessary pieces, moving toward the goal of reducing the total resistance of the path junction to cooler. A good example might be Apple’s MacBook Pro precision aluminum unibody enclosure acting as a heat sink.
DAVID: That’s right and a current example of a good success of this kind of approach would be the Apple G5, Liquid Cooled version.  These are the models with dual processors with a self contained liquid to air cooling. They control the entire system design including the thermal subsystem and have optimized it as a result.
In some sense, this goes back to the put a bigger fan on it mentality. When Apple implemented their solution, you still need roughly the same amount of airflow per watt, but that seems to be coming down a bit. I saw a graph a few months ago that showed that a few years ago typical server air flow requirement per unit of power dissipated was 120 CFM per kW. That has been declining to more like 80 CFM in recent generations. We have to cool more with less airflow. As anyone who’s designed a heat sink knows, your theoretical limit is reached when you hit the specific heat of the air times the mass flow rate. There is no way you can get a thermal resistance lower than that magic number. And sure the cost of going from a 70% effective heat sink to an 85% might be a doubling in the pressure drop for that little gain. That little 15% improvement in the convective part of the HS which might only be in turn 1/3 of the total thermal resistance path. So you find you need to double your fan energy to get say a 5% improvement in cooling. So you cant follow that path very far.
ATS: What do you hope the attendees take away from your keynote at MEPTEC?
DAVID: Within our profession of electronics cooling, I’d like to see my colleagues actually question the temperature specification and ask for a line not a point; can you give me a figure of merit vs. a temperature point. Thermal engineers, in return, can then offer a range of thermal solutions, which all have their particular capital costs, energy, space, and noise requirements associated with them. That way, design teams can choose where they want to be rather then specifying what is often an inefficient place to be.
One rule of thumb is at 11.4 cents a kW hour (about an average world wide commercial electricity cost); a continuous watt cost $1 per year. That gives you a starting point to compare whether it’s worth implementing one or another thermal management strategy. Its based on total cost of ownership (TCO) then. For example, if you want to justify spending a few dollars more on a heat sink, and you have a product life time of several years, you better save at least a watt or two of fan energy to do so. Unfortunately, in all but maybe the datacenter market and a few other markets where you have a person in control of both the energy budget and the capital budget, there is not a great deal of ownership of this TCO. The average guy going to buy a new TV doesn’t figure out in his head as he is walking into the store if his LCD TV is more efficient than his plasma TV is and how many watts of energy that might be and how different his electric bill will be when comparing the two. Many consumer products electrical costs are rarely considered by the end user. The bottom line CAN be calculated but you have to do it.
This kind of TCO analysis is becoming the dominant practice in the server industry where now they are finally considering the big picture of data center infrastructure trade offs vs. server energy consumption. For example, most servers have their fans start to speed up at about 25 degree C in fact that’s a universal specification. So running a data center hotter than that will lead to greater energy consumption from the servers themselves in the datacenter. While there will be savings in the energy consumption of the cooling system of that datacenter. And you have to get down to the specifics of understanding the outside weather even to really deduce the optimum point for your datacenter.
You might be doing a little bit of damage by using this power usage effectiveness metric (PUE), which is the ratio of total data center energy consumption to that of the IT equipment. So, if you start running the fans hotter and the processors need to consume more power due to leakage, we are still talking about IT equipment. Even though you are in a situation where total power consumption is increasing, your ratio still looks better. So its important not to get too hung up on the intermediate measurement rather than the total measurement.
ATS:We’re just about out of time, can you tell our readers what you think of MEPTEC this year?
DAVID: Combining MEPTEC at the same time as SEMI-THERM is an excellent idea. MEPTEC has historically been oriented towards the people who supply the packaging industry, particularly at the device level. SEMI-THERM has traditionally been a lot more at the system level. Those boundaries between levels of packaging and ranges of ownership are a lot less rigid than they used to be in the old days.So its good to get people from different areas talking together.
Be sure to catch David’s keynote, Energy Reduction and Performance Maximization Through Improved Cooling at MEPTEC’s “The Heat is On”, Monday, March 21.