High frequency graphene transistors: can a beauty become a cash cow?

The appearance of graphene to the world of solid-state electronics in 2004 [1] triggered lots of expectations and dreams for a future revolution in micro-electronics. As for nearly any new semiconductor material, for graphene the replacement of silicon was the number one vision especially in the very beginning; this was an expectation that was more based on wishful thinking than on real facts. After realizing that again silicon is unlikely to be replaced by graphene in mainstream electronics at least in the near- and mid-term future, the main hope for graphene-based electronic devices lies on applications in analog high-frequency devices. For these applications the situation is completely different compared to the competition with silicon CMOS. Transistors, the heart of HF electronics, only get very fast for short gate and channel length and with a channel material having high mobility and high saturation velocity. Keeping this in mind, graphene can be considered as the perfect channel material for high-frequency transistors: graphene has the highest carrier mobility and highest saturation velocity of any semiconductor material so far [2–4]. In addition, being only one atom thin, graphene is the incarnation of an ultra-thin-body material and hence predestined for realizing ultra-short channel devices.

From a manufacturing point of view there seems to be no show stopper for the success of graphene in electronic applications. Graphene is a planar material and therefore well compatible with the planar processing technology used for semiconductors. In addition, the synthesis of graphene has been demonstrated on a square-meter scale [5], exceeding the size of Si and III/V semiconductor wafers. One key advantage of graphene might be the flexibility in terms of substrate choice, as graphene can be transferred to nearly any handling substrate ranging from standard Si wafers to PET-foil. Nevertheless, the homogeneity and reproducibility of large-scale graphene growth and especially transfer are still demanding issues, which must be solved in order to meet the semiconductor industry's high requests on device yield.

So overall, having a look at the outstanding intrinsic properties of graphene together with its promising manufacturability, it seems that graphene is the perfect material for high-frequency electronic devices.

Theoretical simulations have quickly confirmed the speed advantages of graphene-based field effect transistors, and operation frequencies up to a couple of THz have been predicted [6, 7], which is well beyond the possibilities of transistors made out of Si, GaAs or InP. Furthermore higher and higher cut-off frequencies, f_T, one key speed indicator for transistors, have been extracted from real devices. For this parameter the current record value of graphene field effect transistors is 427 GHz [8], which is beyond the values achieved in silicon transistors and comparable to the best III/V transistors at similar gate length [9]. A further increase of the speed of GFETs is still realistic. However, for use in HF circuits, speed is not the one and only criterion; functionality matters, too (maybe even more). For analog applications gain is the main functionality of a transistor. Getting voltage gain is especially difficult in GFETs. While in micron-scale devices an intrinsic voltage gain larger than 20 dB is possible and has been demonstrated [10], in GFETs with gate length of 100 nm or less, voltage gain is of order unity or even smaller; hence these transistors are not useful for most analog applications. The absence of voltage gain comes because the drain current in GFETs does not saturate, a behavior which is often explained by the lack of a band gap in graphene. But there might also be a way to get sufficient output saturation without a band gap, specifically due to velocity saturation of the charge carriers, which only can be achieved in high-quality devices.

Moving from individual devices to integrated circuits was the main paradigm change for electronics in the 1950s and the foundation of today's information and communication technology. For graphene-based integrated electronic circuits we are just at the beginning, with only a few demonstrations at GHz frequencies so far [11, 12]. This early stage is not surprising keeping in mind that the functionality of GFETs is still limited. In addition, reproducibility and yield, which are not so important for single devices, where most researchers show only the best-performing ones, matter for integrated circuits (ICs). ICs require a predictable and therefore reproducible device characteristic and a certain yield, as all transistors of an IC must work properly. In terms of yield and reproducibility there is still a long way to go for GFETs and researchers are starting to address these issues.

In the near future the major challenges to be solved will still be on the single-device level in order to increase the device functionality. Achieving gain in short-channel devices will be one decisive issue. This requires a lot of work on the process technology. First, proper output saturation needs to be achieved in this zero-band-gap material. This should be possible using velocity saturation, which would require high-quality graphene material, as given by hBN/graphene/hBN heterostructures. As an alternative or additional option the introduction of a band-gap, as it has been shown for bilayer graphene in a perpendicular field, also improves the output saturation [13]. The reduction of the contact resistance from the metal to the graphene is another important issue for achieving gain, especially for short-channel devices. The recent demonstration of low resistive edge contacts might be a very promising direction to go to further minimize the contact resistance below 50 Ωμm. Values so low are needed because for a 50 nm long device, the graphene channel has a resistance of approximately 100 Ωμm. At high frequencies gain also requires a low resistance of the metal gate finger, a parameter which has not been considered extensively for GFETs. However, because this issue has been solved for conventional semiconductor transistors, e.g. by using T-shape gates, this should not be a major problem at the moment.

The work towards a manufacturable microwave integrated circuit technology is now directed toward creating a suitable wafer-scale technology for GFETs on a silicon or silicon carbide platform. For the case of silicon, the graphene layer is grown on some other substrate, such as copper, and then transferred to the silicon substrate. For silicon carbide, graphene can be grown directly by CVD or by thermal decomposition. It is in any case of utmost importance that the graphene layer is uniform under the transistor, which is typically 10–100 nm long and a few μm wide. In addition, GFETs have shown to have hysteresis in their I–V characteristic, a feature which is probably related to the quality of the gate oxide. The hysteresis effect should be minimized in the near future, since it makes the working condition for the transistor unstable. In the Graphene Flagship, a silicon carbide-based MMIC process has been established using GFETs with f_t/f_max of 20/15 GHz as shown in figure 1. The key target within a few years is to demonstrate receiver and transmitter circuits including amplifiers, mixers and frequency converters using GFETs with significantly increased f_t/f_max.

Zoom In Zoom Out Reset image size

Furthermore, the circuit design also needs to be oriented on the special characteristic of GFETs. So far most ICs containing GFETs use a conventional design, not taking into account the special properties of GFETs, like the possibility of operating a GFET in both polarities and switching from n- to p-type during operation. Also, important parameters for circuits like noise or linearity need to be studied more intensively to give a clear assessment of the performance. And finally, the whole fabrication technology needs to be optimized significantly to get a sufficient reproducibility of single-device parameters. This especially requires work on the growth of graphene and its transfer to the target substrate.

The intrinsic properties of graphene are excellent for realizing high-frequency transistors and the performance achieved so far on a single-device level can be considered quite outstanding, especially keeping in mind that the field is only one decade old. Nevertheless, there are still some major problems to be solved, especially regarding the functionality, reliability and yield. At the moment it does not seem that there are intrinsic roadblocks, but the way will be long and hard, leaving it unclear for a long time if GFETs will become a cash cow or are just a flash in the pan. The real and key advantage of GFETs compared to competing technologies is probably the extreme flexibility with which they can be integrated into nearly any platform. This flexibility is not only unique for a crystalline semiconductor, but might be the key enabler for applications we currently do not even have in mind.

Financial support by the European Commission under contract no. 604391 (Project 'GRAPHENE-FLAGSHIP') is acknowledged.