III-Vs squeeze the terahertz gap
It s the simple goals that tend to capture our imagination, such as the breaking of the four-minute mile and the first ascent of Everest. In our community we also have a similar goal – the building of the first semiconductor chip that can operate at 1 THz. This target is particularly interesting because it is being attacked on two fronts by strikingly different technologies. Quantum cascade lasers (QCLs) are making gains from the optical side as researchers build new structures to operate at longer wavelengths, while engineers are moving closer from the electronics domain with ever faster transistors.
In recent years, advances in transistor speed have been led by Milton Feng from the University of Illinois at Urbana Champaign (UIUC), who has raised his HBT s top speed to almost 800 GHz at room temperature. But he was pipped at the post in December when Richard Lai from Northrop Grumman Space Technology announced that his team s InP HEMTs had passed 1 THz.
Lai, the leader of the company s microelectronics products, reveals that his team has been working under the radar for the last few years, admitting: "This result is not something we focused on to achieve as a goal." Instead, the researchers have been aiming for the targets of a Sub-millimeter Wave Imaging Focal-plane Technology (SWIFT) program, funded by the US Defense Advanced Research Projects Agency (DARPA).
Northrop Grumman s role in this project is to build an amplifier operating at 340 GHz. But to provide amplification at this frequency, you need to produce transistors that are two to three times as fast, which explains why this team has been developing such high-speed devices. The researchers concentrated on characterizing the amplifier and only went on to measure their HEMT s speed when DARPA s program manager, Mark Rosker, asked if the researchers could claim a terahertz transistor. Measurements showed that this was the case and Lai was told to write this up and present the results, which he did at last December s International Electron Devices Meeting (IEDM).
The record-breaking speeds are the result of several modifications to Northrop Grumman s previous HEMT designs. Gate sizes have shrunk to 35 nm and adjustments have been made to the epistructure, including a switch to an InAs channel with very high mobility.
Northrop Grumman s approach to measuring the speed of its HEMTs reflects its target application – microwave devices. Here the key figure of merit is fmax, the frequency that produces a unitary power gain. Measuring this frequency with reasonable accuracy is a tricky task because it involves the extrapolation of data obtained at far lower frequencies (see box "Determining the speed of ultrafast transistors"). Lai accepts that extrapolation-based methods can undermine claims of terahertz records, but he counters this criticism by saying that all of the currently accepted methods for testing produce an fmax of more than 1 THz for his team s transistor. Applying the unilateral gain technique produced a value of 1.2 THz, and figures of 1.1 and 1.4 THz, respectively, were obtained by extrapolating maximum stable gain and making circuit model calculations.
Lai claims that the ultimate validation for fmax extrapolation above 1 THz comes from the performance of the team s three-stage low-noise amplifiers that have also been built from these HEMTs. At IEDM he unveiled the results for a 0.65 × 0.35 mm MMIC chip, which can produce 21 dB of gain at 285 GHz, 18 dB at 300 GHz and 15 dB at 340 GHz. And he believes that amplifiers operating at much higher frequencies are also within his grasp. "We have devices with 1.4 THz fmax, so you should have some decent gain at 600 and 700 GHz. This is not our primary thrust but we are working on it." At these frequencies the difficulties are not associated with the chip itself but with the coupling of signals in and out of the transistors. Simulations have shown that this can be done, but his team is still to follow this up with a real demonstration.
Lai is also very keen to emphasize that his results are not just one-offs. "In industry my objective is not to try to demonstrate one terahertz transistor. This has to be a technology that can be manufactured, so yield and reliability are very important." Northrop Grumman has already produced tens to hundreds of wafers with this process and yields are encouraging. In particular, the process is producing T-gates with 99% yields that are also robust, according to temperature cycling and vibration tests.
One of Northrop Grumman s next steps will be the transition from prototype fabrication to genuine deployment. "We ve got to get to the next level where we can make [our HEMTs] into blocks and start putting things together. We re still at the fundamental technology stage, trying to get the basic components and understand them."
Lai believes that the shipments of modules employing high-speed HEMTs for military applications, such as radar, will start on a very small scale. However, a lot of arrays will need to be built, and this could translate into substantial chip volumes.
From HEMTs to HBTs
Feng, meanwhile, has a very different application in mind for his high-speed HBTs – mixed circuits such as analog-to-digital converters. This application has a different key figure of merit, fT, the frequency at which current gain is equal to unity. (However, a high fmax is needed for a wide dynamic range.) At IEDM, Feng s team reported a speed of 683 GHz for its double-heterostructure HBT (DHBT), which increased to 745 GHz at –37 °C. "The device is not optimized," said Feng, "and we are expecting something even better to come out soon." He hopes that this structure can operate at more than 1 THz and believes that he can achieve that this year.
Up until very recently, many of Feng s IEDM papers focused on the development of conventional single-heterostructure HBTs, but increasing the speed of these transistors has come at the penalty of a very low breakdown voltage. These devices could achieve terahertz operation at a marginal breakdown voltage, says Feng, but this is not good enough for mixed-signal applications. "In order to achieve [a breakdown voltage of] about 2 V in the terahertz regime, the technology requires the collector to be a larger bandgap," he added.
Switching to DHBT structures allows just this, and the UIUC team has been developing a type-II design that features a pure InP collector layer (figure 1). Feng claims that this is a superior structure to type-I versions, which have a transition layer in the collector section that causes current blocking. In comparison, type-II structures have a base layer with an energy band above the collector, which gives an advantage known as velocity overshoot. "Fundamentally, that gives you a fast transport through the collector," said Feng. His type-II designs also deliver better thermal conductivity, thanks to binary material in the collector, which is said to improve I-V characteristics and allow the HBT to operate at lower temperatures.
Like Lai and his team at Northrop Grumman, if Feng wants to make circuits that can operate close to 1 THz, he will have to build HBTs with top speeds that are way beyond that value. However, he thinks that will be possible with his current structure. "The problem right now is that we have a 400 or 350 nm emitter. If you look at silicon, it s 130 or 65 nm, but if we can get to that level we will cross over into the terahertz range."
One of the major players operating on the other side of the terahertz gap, and hoping to come down in frequency, is Jérôme Faist from the University of Neuchâtel, Switzerland. He and his team currently hold the record for the lowest frequency for a QCL, which stands at 1.2 THz for a GaAs/Al0.1Ga0.9As-based heterostructure. However, this laser can emit at just 850 GHz when a strong magnetic field is applied with a superconducting solenoid.
The emission from this type of device results from transitions between extended states that are similar to those found in superlattices, and growing such a structure is not easy. "If you were just to scale linearly all of the dimensions with wavelength, you would end up having unreasonably thick layers," explained Faist. He says that this can be overcome by using metal-metal waveguides, a common solution for many researchers in this field: "It s conceptually simple. There is some [process] know-how that you need to acquire, but in the end it turns out to be quite straightforward."
Designs must also combat potential losses from free-carrier absorption. This absorption process is predicted to be proportional to the square of the emission wavelength, but Faist found that the loss did not follow this trend and only showed a small increase between 2 and 1.2 THz. "It s a good surprise and it gives hope that we can proceed [to shorter wavelengths]."
Avoiding the problems associated with free carrier absorption has not been down to good luck and is the outcome of work with theoretical models that consider free-carrier absorption as a tail of intersubband absorption. "If it s intersubband absorption, you can t completely eliminate it, but you can push it to the places where you can tolerate it and remove it from the places where you don t want it."
One of the major weaknesses of all terahertz QCLs is their low operating temperature. The University of Neuchâtel s 1.2 THz QCL, for example, can operate in pulsed mode at temperatures of up to 69 K, but it can only produce a continuous-wave output at up to 50 K. However, Faist believes that progress can be made by switching the quantum well from GaAs to InGaAs. "We will benefit from a slightly lower optical phonon scattering rate, and since optical phonons are the main scattering mechanism at high temperature, this should really help us."
Friends or foe?
If we assume that transistor and QCL technologies both continue to extend their spectral coverage, then there could come a day when you have a choice of device for a terahertz source. Feng is adamant that at that stage the transistor will win because it operates at room temperature and doesn t require cooling. However, Faist believes that a marriage of the two technologies could lead to even better performance characteristics.
"One of the problems that we have with QCLs is tuning," explained Faist, who says that these devices actually have the potential to tune over a broad range, thanks to their wide gain bandwidth. Electronic components can easily provide a tunable source, however, and Faist believes that using this device in combination with a QCL providing amplification would be a great solution.
Lai is also enthusiastic about the possibility of uniting the two technologies. "It could probably be a hybrid solution in the end, depending on what you re trying to do." And it definitely looks like Northrop Grumman is keen to invest in this new field, as the company is planning to set up a lab dedicated specifically to terahertz research.
