Semiconductor manufacturing techniques bring benefits to thermal management

"Only a few spots on a chip run hot," he claimed, "and the traditional thermal management approach is to cool the whole chip. This not only affects the chip's performance, it also throws heat out into the room and the process is inefficient." Nextreme's solution is to address the problem at the micro level. "We can cool the chip right at the point where the heat source is," Magill continued. And it does that using the Peltier effect, implementing it using thin film thermoelectric technology and what the company calls thermal bumps. Nextreme is a four year old company which spun out from the Research Triangle Institute in North Carolina. "We have a strong IP position," Magill claimed, "with around 30 patents granted or applied for. The two important ones, however, relate to our process technology." But it's not all Nextreme's work; it has also licensed thin film thermoelectric technology from NASA's JPL division. But, as Magill admits, there's nothing new in thermoelectric technology. "It's been around for about 60years," he noted, "and, other than some performance improvements, nothing has really changed." Historically, the technology has found application in telecoms and photonics. "If you look at these sectors," Magill continued, "the drive has been towards miniaturisation, but thermoelectric coolers haven't shrunk; they are quite large and expensive. But thin film technology allows us to miniaturise the devices." He believes this is a similar step function to the use of silicon in integrated circuits. "We hope this platform will give us the same opportunities," he noted. While Nextreme's first products – shipping since January 2008 – were targeted at photonics applications, products are now being developed for spot cooling on chips. To this end, Nextreme has been engaged on a project with Intel to explore the possibilities. "When you look at the electronics world," Magill explained, "it's all about material deposition; thermal management is all about material deposition, but electronics companies are not used to putting thermoelectric coolers into their products. So we're looking at how these can be better integrated into chips." According to Magill, the project had developed one way of solving the problem when they realised that the technology was being integrated into flip chip packaging. "So we continued in that direction," he observed. Working with Intel has brought the company access to what he described as 'mainstream' packaging technology. He said that Nextreme had realised there was a significant amount of redundancy in flip chip packaging. "Particularly in power and ground," he added. "There may be 7000 bumps on the package, but only 300 may be used for signals; the rest are redundant. We can cool these chips by putting just 10 or 20 thermal bumps on the die and these may only be 250µm in diameter. Our long term vision is that designers will find out where hot spots might occur and include thermoelectric bumps." Nextreme is pursuing this route even though Magill admits that thermoelectric cooling has a 'bad name' for high power consumption. "This is because people have used the technology near ?t max. If you keep ?t within a certain range, you only need to cool 8 or 10°, not 50 to 70°." Noting that thermal management is developing to be a 'universal problem', Magill said larger form factors can't be used any longer. "However, unless you design our approach in from the start, it's not a drop in solution, so mindsets have to change." Typically, Nextreme's thermoelectric coolers are between 5 and 25µm thick and are created using familiar semiconductor manufacturing techniques. "There are two particles which carry heat," Magill explained. "Phonons – if they exist – are quantised lattice vibrations, but are not directional and tend to 'smear' heat. The other particle is the electron, and we can use that in some materials to carry heat." The attraction of semiconducting materials is that heat flow can be caused to be directional. "Electrons carry heat from one surface to another," Magill continued, "but a return path is needed. This requires two types of material: one where the main carrier is electrons; the other where the main carrier is holes. So we need p and n type materials." The process starts with a thin film of bismuth telluride, which is implanted with different materials to create the p and n type carriers. "The thin film is in the form of a single crystal," Magill added, "and we're doping it with donors and receptors." In order to compare thin film coolers with other approaches, Nextreme uses heat density as a metric. "Block thermoelectric coolers typically pump 10 to 13W/cm2," Magill claimed. "But we have devices that can do up to 120W/cm2. It's an order of magnitude improvement because it's an inverse ratio to material thickness." Amongst Nextreme's range of discrete devices is the UPF40 thermoelectric module, aimed at cooling and temperature control of optoelectronic devices. Part of the OptoCooler family, the module can achieve a heat density of 72W/cm2 at 25°C, removing 3.7W of heat using an active footprint of 5.1mm2. And the latest device is the OptoCooler HV14, described as the first module in a new class of high voltage and high heat pumping thermoelectric coolers that operate at low currents and are optimised for standard circuitry and power requirements. According to Nextreme, the device can pump up to 1.5W of heat at 85°C and operates at a maximum of 2.7V with a maximum current of around 1A. With a footprint of 2.8mm2, the module is said to be suited to cooling and temperature control of optoelectronic devices. Magill noted a limitation of the technology. "The larger the bump, the larger the current needed to drive it," he said, "so there may well be some physical limit that we will encounter. "As we shrink the bumps, the current needed to drive them also shrinks. At a bump size of 100µm, we have lower current and higher voltage, but we get the same amount of cooling." Nextreme's technology can be applied in a number of ways; not only can the technology support cooling, it can also work in reverse. Using the Seebeck effect, waste heat can be turned into electricity, generating more than 3W/cm2. Apart from developing discrete cooling and energy recovery devices, Nextreme is also offering licences for integration of the approach at the die level and at the wafer level. "Our technology can scale to any size," Magill concluded.