Over the past decade, there has been an explosion of interest in ferroelectric hafnium oxide (HfO2) and in devices made from it. This has been further fueled by the relatively simple fabrication and the pervasive use of the material is the semiconductor industry. A device that is attracting particular attention is the ferroelectric field-effect transistor (FeFET), which contains a thin HfO2 layer in the gate stack and can display at least two stable conduction states, thus holding promise for non-volatile memory applications. Moreover, some of its peculiar electrical properties, which closely resemble the behavior of biological neurons and synapses, make these devices interesting for the brain-inspired neuromorphic computing.
Our group at NaMLab, in collaboration with our industrial partner GLOBALFOUNDRIES and with Fraunhofer IPMS-CNT, has been exploring HfO2−based FeFETs since they were first reported in 2011. Since then, rapid progress has been achieved, including the scaling to small device dimensions and integrated large-array demonstrators. Nevertheless, as any new technology, FeFETs have their impressive strengths and but also weaknesses. While the former ones are typically obvious and tend to be emphasized, the latter ones tend to be much less obvious, thus deserving a careful investigation.
Last year we began to investigate in more detail different leakage sources that may be undesirable in FeFET arrays. In particular, we explored the gate-induced drain leakage (GIDL), which is a tunneling current taking place at the gate-drain overlap of a transistor. At the moment, the GIDL is one of the main leakage sources in modern scaled devices and many strategies are being developed to minimize its impact. As expected, the GIDL current in FeFETs could be largely influenced by several structural changes; for example, by varying the layer thicknesses in the gate stack, the overlap extension between the gate and drain regions as well as the doping of the drain region. Unexpectedly, however, we have observed that the transfer characteristics (ID-VG), which is composed of the GIDL branch and of the regular subthreshold-inversion branch, becomes increasingly symmetrical for some specific sets of structural and electrical parameters. Additionally, under a very particular combination of fabrication steps, we achieved the GIDL current to be substantially independent of the ferroelectric polarization, which is typically not the case.
This unexpected symmetry triggered a completely new idea of FeFETs as frequency multipliers. In fact, already in the 1980’s there have been proposals to improve the efficiency of frequency multipliers by exploiting symmetries in the I-V curves of electron devices. While the antisymmetric I-V curves tend to enhance the generation of the odd-order harmonics, the symmetric I-V curves promote the even-order ones. Over the course of the last year, we worked to find those electrical conditions that best stabilize the symmetric parabolic I-V curve in FeFETs, both by tuning the GIDL current and the ferroelectric switching, as explained in our manuscript and shown in Fig. 1a. We then performed experiments, where a sinusoidal signal Vin is applied to the gate terminal, which is dc biased at the inflection point of the parabola (Fig. 1b). As a result, an output signal with the frequency twice that at the input could be successfully obtained, thus confirming the frequency doubling (Fig. 1c). This encouraged us to carry out many additional tests that indeed confirmed our predictions and proved some further properties/functionalities, such as frequency modulation, insensitivity to endurance degradation etc.
At this point we would like to stress the importance and constructiveness of the peer review process. In fact, the reviewers requested additional experiments regarding the extension of the operating frequency range and the scaling capabilities. Although this has meant additional time and efforts, the obtained results have largely improved the quality and potential impact of the paper, for which we are very grateful.
Interestingly, this manuscript shows one of those examples of how a harmful device property, such as the GIDL, can be turned to advantage. Nevertheless, there is still a lot of room for improvements for the FeFET-based frequency multipliers, as we outlined in the conclusions of the manuscript. The HfO2-FeFET is still in its infancy and we expect that further material and technology advances may bring it closer to this exciting application.
If you are interested in learning more about this topic, you can read our article "Reconfigurable frequency multiplication with a ferroelectric transistor" at https://www.nature.com/article...
Fig. 1 Frequency doubling with a FeFET: (a) Tuning of the transfer characteristics by varying the GIDL current; (b) Electrical set-up for frequency doubling; (c) Experimental result of an output signal with frequency twice that of the input signal.