Designing for extreme temperatures

6 mins read

For decades, electronics design has relied on the temperatures inside a system never getting too far away from room temperature, even when those devices are specified for military applications. In practice, few devices are ever expected to work at -55°C or more than 125°C even; in fact, many fail to function below freezing.

A lot of the properties of CMOS improve at low temperatures, such as saturation current and transconductance – so devices should be able to switch more quickly. KryoTech took advantage of that several years ago by building cooling chambers for PCs based on the AMD Athlon processor that operated the CPU at an ambient temperature of –40°C so the processors could be heavily overclocked. Yet many devices will fail as the temperature drops. The problem is that as the system gets colder the threshold voltage shifts and hot carriers cause greater levels of damage. Although devices should be faster, the trick is to get them to switch state at all. Increasing the operating voltage helps CMOS function at very low temperature but, at the same time, further increases the risk of damage. Instead of silicon CMOS, which can be made to operate down to around –230°C (40K), researchers have proposed that cryogenic devices be used to increase compute throughput massively – using device architectures that could take advantage of the lack of thermal noise in circuits cooled below the temperature of liquid nitrogen. However, no-one has realistically wanted to invest heavily in these technologies because of the high cost of refrigeration and a lack of evidence that the circuits would not simply be overtaken by the high pace of development in CMOS. So, the superconducting Josephson junction and the rapid single flux quantum logic technology on which it is based remain largely intellectual curiosities. Today's experimental quantum computers – a class of machine that could ultimately replace conventional architectures – rely on extremely low temperatures, but researchers hope they can, as with supercomputers, develop systems that will work at or near room temperature. At the other end of the scale, excessive temperature is usually just bad for circuitry. Thermal runaway is a concern for modern CMOS processes. Leakage worsens as a circuit gets hotter until the device reaches the point where it simply fails to switch. Bipolar devices fail simply because of the way they work: at more than 200°C, charge carriers are produced spontaneously as the temperature rises, so the transistor functions purely as a conductor. The space industry presents the greatest demands to electronics when it comes to temperature compatibility. For a probe to go to Neptune, it needs to be able to operate in temperatures as low as –230°C. A lander that can operate on Venus would have to tolerate temperatures of close to 500°C. Very low temperatures are not just seen near the outer planets. Simply being in shadow in orbit around the Earth or at one of the lunar poles is enough to have temperatures plunge to –230°C, well below the temperature at which liquid nitrogen boils. NASA has busily been experimenting with a variety of techniques to deal with extreme temperatures. Traditionally, the agency's subcontractors have selected the simplest option: use heaters and coolers to keep the temperature within the range of conventional electronics. A large interplanetary probe, such as Cassini, may have close to 150 individual radioisotope heaters, each one the size of a roll of coins. Ultimately, NASA wants to dispense with the heaters as they increase cost, weight and complexity. However, heating can cause design problems. The James Webb infrared telescope, which is meant to go into Earth orbit before the middle of the decade, will use a sunshield that will keep one side of the probe in a permanent freeze while the other will be side will be hot. Infrared sensors tend to depend on very low temperatures to operate. Putting a heater next door will help some of the electronics, but destroy the performance of those sensors. Operating in cold or hot conditions is not the only problem. Sudden changes in temperature will affect Moon and Mars landers as well as satellites as they move out of shadow and into direct sunlight. The temperature cycling could cause bond wires to break suddenly and damage packages, particularly as solder generally has a much lower melting point that any semiconductor device. As CMOS, with adequate protection, shows promise at cryogenic temperatures, it looks to be less of a problem. Silicon based bipolar devices, however, fail at around –195°C and higher mobility materials are needed to get around this. The focus in space electronics is currently on silicon germanium, although gallium arsenide or germanium are contenders. Closer to Earth, the demands for low temperature operating tend to be for very sensitive analogue circuits or for medical imaging, where the circuits are used to control the supercool magnets employed in CT scanners. A growing number of applications are putting demands on semiconductor devices at the hot end of the range. In mining and oil exploration, automotive parts maker Bosch is among those working on chips that will work at temperatures in excess of 200°C. The aim is to get microcontrollers into the lubrication oil used to cool the transmission in motor vehicles so the control electronics can form part of the mechanical parts. That oil can get as hot at 150°C and the electronics will run hotter than that. Recently, cars started shipping in which the controllers are mounted next to the actuators, which has pushed the junction temperature up to 175°C. When the electronics are integrated within the lubricated actuators, the junction temperature will rise beyond 200°C. Similarly, the power transistors used in electric vehicles will need to run at elevated temperatures to avoid the need for energy sapping forced air cooling. Despite the high temperatures, Bosch is continuing to work with standard silicon processes. In principle, altering the circuit and transistor design should allow the devices to work reliably at elevated temperatures and deal with thousands of temperature cycles – dropping below 0°C when the car is stopped, then rising to more than 200°C. For higher temperature semiconductors, silicon on insulator (SOI) technology shows promise. The thin silicon sandwich limits the effects of unwanted thermal charge carriers that afflict bulk processes. Suppliers such as Belgium based Cissoid have made logic devices that will function at more than 200°C and research indicates that SOI can get to at least 250°C. Some of the properties of SOI improve when it gets hot. Research on high temperature SOI performed by Cissoid and Oki Electric, another SOI specialist, found some peculiarities in the behaviour of the material at high temperatures. One thing they found was in the gate induced kink effect of thin body SOI devices. This normally appears as a large peak in transconductance versus gate voltage at less than the maximum gate voltage. Although prevalent at room temperature, it pretty much disappears at higher temperatures. Noise can be problematic in SOI, but modelling work has indicated that 1/f noise is not tied directly to temperature. However, noise is worse in partially depleted SOI rather than the fully depleted form, which has a much thinner silicon based channel. The fully depleted form fares better in terms of noise and this is good news for high temperature operation as thin body devices become more readily depleted as they get hotter. For power transistors and very high temperatures, wide band gap materials, such as diamond and silicon carbide, show the most promise. Researchers from NASA's Glenn Research Centre have demonstrated an amplifier based on silicon carbide that could function at 500°C, glowing red hot but still functional. Placed on a ceramic carrier and coupled with aluminium oxide capacitors, the amplifier ran stably during tests lasting more than 400 hours, indicating the team was not just getting close to semiconductors that could survive the horrific conditions on planets such as Venus, but that the packaging technology could as well. Silicon carbide has even presented the possibility of a new type of transistor that can deliver extremely low leakage at temperatures higher than 300°C. Developed at Case Western Reserve University, the design exploits electromechanical switching – it is, in effect, a miniature relay. Although it relies on the movement of a cantilever under an electric field, the team demonstrated switching at up to 1GHz in temperatures approaching 500°C. Circuit design is not very different to conventional CMOS – an inverter uses complementary pull up and pull down stages analogous to n- and p-channel transistors. At such high temperatures, oxidation is a problem, so the devices had to be held in vacuum conditions inside ceramic packages. Diamond has the added benefits of a very high thermal conductivity, a large bandgap only matched by aluminium nitride – suiting it to power devices – and an electron mobility similar to that of gallium arsenide. The trouble with diamond is whether you can make it in the first place. It is possible to synthesise diamond using chemical vapour deposition (CVD), but this process tends to result in poorly performing polycrystalline material. However, researchers have managed to control the deposition closely enough to make monocrystalline layers. Because of its thermal conductivity, diamond could form a key part of packaging materials for hot electronics. One proposal from the University of Bristol is to make a heat transfer baseplate for high power gallium nitride transistors out of a silver composite that contains tiny diamond crystals. However, as silicon itself has an extremely high melting point, the focus remains on the most commonly used semiconductor material, and its alloys, as the basis for extreme temperature electronics.