Ferroelectric gate on AlGaN/GaN heterostructures

The capability of switching the spontaneous polarisation under an applied electric field in ferroelectric materials can be exploited for the use in low power, non-volatile, re-writable memory devices. Currently available commercially is ferroelectric random access memory, FeRAM, which allows for high speed, low voltage and greater write-erase endurance compared to its two main competitors, Flash and EEPROM, when using the one transistor – one capacitor configuration. However, it is desired to further optimise the configuration in order to obtain better densification, faster access time and better reliability. One way to do such is to pass from the ferroelectric capacitors and develop ferroelectric field effect transistors. Exploiting the phenomenon of ferroelectricity and integrating ferroelectrics with the semiconductor technology has not been simple. The FeFET has been demonstrated using a silicon-based transistor, however commercial devices are not available. Challenges arise mainly due to the high temperature deposition of perovskite ferroelectrics causing the degradation of the ferroelectric/semiconductor interface due to inter-diffusion. Acquiring long term retention of the transistor behavior has also been problematic due to phenomenons such as charge injection and depolarisation. In this thesis a new approach to the problem of semiconductor devices with a ferroelectric gate is explored. Instead of using a silicon-based device, semiconductor heterostructures are investigated. Combining the high mobility channel existing in semiconductor heterostructures, with the non-volatile switching of the polarisation in the ferroelectric gate can pave the way to novel future devices. The AlGaN/GaN semiconductor heterostructure was chosen for two main reasons. The first is a two dimensional electron gas, 2DEG, located at the AlGaN/GaN interface which possesses better transport properties than a single layered semiconductor. Secondly, GaN and its alloys are known to have large chemical and temperature stability making them ideal to withstand the high temperature deposition process of perovskite ferroelectrics. The deposition of two ferroelectric layers onto the AlGaN heterostructure were investigated. Lead zirconium titanate, PZT, a traditional perovskite ferroelectric deposited at high temperature, was chosen for its high remanent polarisation and low coercive field. An alternative ferroelectric gate, the co-polymer poly(vinylidene fluoride/trifluoroethylene), P(VDF/TrFE)(70:30) was deposited and of interest due its low crystallisation temperature and low dielectric constant. Its remanent polarisation is smaller and coercive field larger than that of PZT, but were determined sufficient to observe the depletion effect in the two dimensional electron gas. The goals accomplished in this research were: Development of Ferroelectric Gate Processing: Deposition processes of the ferroelectric layers were developed and optimised in order to obtain a high quality ferroelectric, while maintaining the original transport properties of AlGaNs 2DEG. The processing of HfO2 and MgO buffer layers were developed and investigated for their effects on limiting unwanted inter-diffusion and charge injection. PZT Gate on Al0.3Ga0.7N: The first successful development and observation of a PZT gate on a Al0.3Ga0.7N/GaN heterostrucutre was accomplished. A (111) oriented PZT(40:60) ferroelectric gate on the Al0.3Ga0.7N heterostructure depleted the sheet resistance of the 2DEG by a factor of three when poling the PZT layer with –40V directly to the conductive cantilever used in a piezoresponse force microscope. This decrease in sheet resistance was stable for more than three days. P(VDF/TrFE)(70:30) Gate on Al0.3Ga0.7N: The ferroelectric co-polymer P(VDF/TrFE) (70:30) was investigated as a gate on the Al0.3Ga0.7N heterostructure. When the P(VDF/TrFE) was poled with –30V to a top electrode the sheet resistance was modulated by a factor three, however there was no retention of this modulation. Ferroelectric Spontaneous Polarisation on a Semiconductor Heterostructure: It was theoretically derived that when depositing a ferroelectric layer onto a semiconductor heterostructure its spontaneous polarisation is dramatically reduced due to the depolarisation field and may be suppressed completely. This decay in spontaneous polarisation can be minimised by using a ferroelectric layer with a small dielectric constant, which justifies the use of ferroelectric polymers as the gate material. Domain Writing:The technique of domain writing using piezoresponse force microscopy showed that it was possible to create domain patterning on the ferroelectric layers that were deposited onto the Al0.3Ga0.7N heterostructure with sub-micron line resolution, on the order of 300nm or better. Using ferroelectrics on semiconductors is not only limited to ferroelectric memory devices. Other applications could be ferroelectric nano-lithography, where a ferroelectric pattern is written with a piezoresponse force microscope, PFM, and this pattern is projected onto the transistors channel, potentially creating quantum/ballistic devices. The challenge of integrating ferroelectrics with semiconductors has yet to be fully understood. One of the main advancements that was reached with this research is the understanding of the decrease of the spontaneous polarisation when a ferrroelectric layer is deposited onto a semiconductor heterostructure of finite thickness. The maximisation of the ferroelectrics polarisation on 20nm of Al0.3Ga0.7N can be obtained when minimising the dielectric constant of the ferroelectric layer. The current challenge is to get a better physical understanding of the limitation of the retention of the spontaneous polarisation, while discovering methods for its improvement.

Polarisation reversal in ferroelectric PVDF and PZT films

Roman Gysel

Ferroelectrics are characterised by a spontaneous polarisation that can be reversed by an external electric field. The stability of the polarisation states and the possibility for controlled switching between the states render ferroelectric materials very attractive for memory applications, in which polarisation states are associated to a binary information "0" or "1". Fast, dense, and reliable data storage has gained enormous importance in our information technology based society. In this context, the fundamental physics of ferroelectric materials attracted great interest with the ultimate goal of reaching memory elements with shortest access times and highest densities. Two model systems of ferroelectric materials were studied in this thesis on the polarisation reversal: an inorganic, classical perovskite and an organic polymer. Despite the fundamental structural differences between lead zirconate titanate [Pb(Zr,Ti)O3] and the investigated polyvinylidene fluoride-trifluoroethylene [P(VDF-TrFE)] copolymer, the switching kinetics in these materials can be described by a similar formalism. Our approach of a combination of macroscopic and microscopic techniques enables an insight into the mechanisms of the polarisation reversal and allows for a better understanding of macroscopic effects governed by mechanisms occurring on the nanoscale. The experimental techniques developed for the ferroelectric capacitor structures have been successfully applied for studying a first hybrid PVDF/silicon ferroelectric transistor for non-volatile information storage. The major accomplishments of this thesis are the following: We demonstrated for the first time a piezoelectric scanning probe study on the cross section of ferroelectric thin films. In combination with switching pulses, this allowed us to directly image the polarisation reversal in ferroelectric films quasi-dynamically. The utility of cross-sectional domain imaging as a powerful tool for the study of the polarisation reversal in ferroelectrics has been corroborated. The apparent contradiction between observations interpreted as a fast sideways growth and the classical scenario of a fast forward growth could be explained by an oblique polarisation direction with respect to the film plane. A surface-stimulated nucleation of reverse domains, with preference on the electrode with a negative potential was observed. It was interpreted in terms of a built-in field due to a depletion layer at the ferroelectric-electrode interface. A combination of microscopic and macroscopic methods used for the first time in organic ferroelectric thin films enabled a better insight into polarisation reversal of the copolymer of vinylidene fluoride and trifluoroethylene P(VDF-TrFE). The polarisation reversal was found to be impeded by a restricted domain growth. This limits the applicability of the conventional Kolmogorov-Avrami model as well as the nucleation-limited switching model with a broad distribution of nucleation times. We demonstrated that an interface-adjacent passive layer may impact on the switching properties and gives rise to a retardation of the polarisation reversal. Furthermore, an extraordinary dielectric constant increase was observed in the films with a passive layer due to an additional domain wall contribution. We fabricated for the first time a hybrid silicon/ferroelectric polymer ferroelectric single transistor memory element with non-volatile operation. The observed difference of the polarisation reversal kinetics of the ferroelectric in the simple capacitor and in the transistor structure was explained by the gate oxide buffer layer in the transistor. Our previously developed tools and approaches enabled a better understanding of retention loss mechanism. A semi-quantitative retention model was suggested that explains the observed exponential retention loss. The investigation of relevant fundamental aspects of ferroelectric materials and the development of novel characterisation techniques in this work can be of interest for further development of non-volatile memory devices.

EPFL2008

Patterned and self-assembled ferroelectric nano-structures obtained by epitaxial growth and e-beam lithography

Simon Bühlmann

The continuous downscaling of microelectronic circuits combined with increasing interest in ferroelectric thin films for non-volatile random access memories (FeRAM) is drawing great attention to small ferroelectric thin film structures. There are various challenges and open questions related to processing and theoretical understanding. The processing must assure damage-free ferroelectric capacitors of reproducible properties. More theoretical understanding is necessary to estimate the impact of size effects on the stability of the ferroelectric polarization, domain configurations, switching properties, and other properties such as the piezoelectric response. Pb(Zr,Ti)O3 (PZT) is one of the favorite materials for FeRAMs. Today's capacitors are poly-crystalline with sizes in the micron range. Each capacitor is composed of a large number of grains, guaranteeing the averaging of electrical properties. If only a few grains were present, the scattering of properties from one capacitor to another would take place. This can be avoided using oriented single-crystalline capacitors. An important processing issue is patterning. The switchable polarization might be reduced due to etching damage, a frequently reported current fabrication problem. There is also the question of domain configuration - processing relations. The optimal solution of (001) oriented tetragonal PZT cannot be realized in large capacitors because ferroelastic 90° domains form spontaneously upon cooling through the transition temperature and are stabilized for reasons of stress compensation. In this work, we investigated two routes to fabricate small ferroelectric structures: A subtractive route and an additive route. Patterning was done by electron beam lithography (EBL) in order to access the sub 100nm range. Dry etching was done in an electron cyclotron resonance (ECR/RF) reactor working at very low pressures and ion energy. In the subtractive route, the starting point were continuous 50 to 200nm thick Pb(Zr0.4,Ti0.6)O3 films, which were deposited on conductive, Nb-doped SrTiO3 (100) substrates. The films were c-axis oriented with an a-axis fraction of about 20%. The films were patterned by means of EBL. The positive poly-methyl-methacrylate (PMMA) resist was spun on the PZT film, a dot pattern was directly written by the e-beam, and after development a Cr layer was evaporated. The obtained Cr patterns served as a hard mask for PZT dry etching. Features with an aspect ratio of up to 2 could be cut out of a 200nm thick PZT film. We have shown that dry etching in an electron ECR/RF reactor provides damage-free single crystalline sub-200nm features which were still ferroelectric. The smallest features obtained had a lateral size of 80nm. We found that the resolution of the EBL was limited by the backscattering of electrons from the high density PZT layer. Piezoelectric sensitive atomic force microscopy (PAFM) revealed an increase in the piezoelectric response when the feature's aspect ratio was increased. The measured piezoelectric coefficient increased by a factor of more than three compared to the continuous film. Un-clamping was found to account only for a portion of the increase. As major contribution, we suppose a change in the domain configuration from a to c, and unpinning of domains. This hypothesis was supported by local PAFM measurements on the continuous film, where the same behavior was observed, when the film underwent local training cycles just before the measurement using pulses with voltages well above the coercive field. It is also important to investigate non-destructive fabrication processes to reduce the negative impact of interface defects and grain boundaries. This was achieved using an additive route. The fact, that PZT nucleation on Pt surfaces is difficult but easy on TiO2 was used for the site controlled growth of PZT single crystals on patterned 2nm thick TiO2 layers serving as affinity spots. This route was carried out on Pt (111) and (100) surfaces. A 50nm thick Pt layer was epitaxially grown on single crystalline SrTiO3 (111) and MgO (100) substrates, and then covered by a 2nm thick TiO2 layer. On both substrates, the TiO2 was found to be (100) oriented. In this route, EBL was used to pattern the TiO2 layer into seed islands. As only 2nm of TiO2 layer have to be etched, a new type of negative organic mask could be used, which circumvented the lift off process. Direct deposition of Pb(Zr0.4,Ti0.6)O3 on the seed island covered Pt (111) surfaces led to the nucleation of triangular shaped crystals only on the seeds and not elsewhere. The lateral size of the triangles was between 30 and 150nm. The uniform orientation of the triangle's sidelines implies epitaxial growth of the PZT. Applying a 1nm thick PbTiO3 starting layer prior to PZT deposition increased the nucleation density. A 2nm thick PbTiO3 starting layer prior to PZT deposition led to square shaped crystals on small TiO2 seeds implying the growth of (001) PZT. On Pt, triangular crystals were obtained. For this deposition, the squares showed a uniform size distribution, but their in-plane orientation was random. In all cases, the nucleus density was found to be 60 times less on Pt than on the affinity spots. The nucleation controlled character was expressed in an observable depletion layer around larger TiO2 seeds. A theoretical nucleation model was able to well explain this behavior and the relevant parameters such as activation energies of diffusion and desorption could be derived from experiments. On the Pt (100) surface, direct deposition of PZT led to the nucleation of a non-ferroelectric phase on most parts of the substrate. A depletion around TiO2 seeds was again observable. PZT crystals were found exclusively on some TiO2 seeds. PAFM measurements revealed that all PZT crystals obtained by the additive route were ferroelectric. The ferroelectric features have been investigated by means of PAFM. Complex domain structures were identified, the smallest observed domains had diameters of 15nm. The thinnest investigated feature showing ferroelectricity was 6nm thick. An attempt was made to compare experimental findings with the Landau-Devonshire-Ginzburg phenomenological theory. The d33,f loop together with the CV-loop of a 600nm thick, mostly c-oriented film could well be described by the theory. Deviations could be identified as domain wall contributions. The critical thickness for ferroelectricity was estimated using in addition literature data for domain wall energies. The critical thickness was calculated as 1 to 2nm, which does not contradict experimental findings.

EPFL2004

1T Memory Cell Based on PVDF-TrFE Field Effect Transistor

Didier Bouvet, Roman Gysel, Mihai Adrian Ionescu, Sebastian Riester, Giovanni Antonio Salvatore, Nava Setter, Igor Stolichnov

Interest in PVDF-TrFE copolymers as ferroelectric material for Memory application is driven by the prospect of having low cost, low operating voltage and fully organic device. Some previous studies reported FET designs using copolymers [refs 1,2] but none of these structures were fully integrated on silicon wafers and using a MOSFET fabrication process. We present for the first time the integration of a PVDF-TrFE (70%-30%) layer into a standard n-MOS transistor through a quasi-standard semiconductor technology. This allows us to achieve a Non Volatile Memory cell and at the same time to compact capacitor-transistor ferroelectric cell into a one-transistor memory cell. The ferroelectric polymer is prepared using a new recipe based on Methyl-Ethyl-Ketone, which helps to reduce considerably the film thickness and the coercive field. The solution is deposited by spin coating down to 100nm and the structures are defined by standard UV lithography. We add a 10nm SiO2 layer between the polymer and the p-doped substrate in order to reduce losses and the screening in the ferroelectric gate (Fig.1a). The ferroelectric properties of the polymer are preserved during the whole process (Fig.1b). We design, fabricate and characterize transistors of different dimensions (from 50μm down to 2μm as channel length and width) demonstrating a good scalability. Experimental Id(Vg) and Id(Vd) curves for different channel lengths and widths are in agreement with the theoretical predictions for such a device (Id~(W/L)Vd. Fig.2-3). The device shows very good hysteresis properties translated in a controllable and reproducible memory behavior. The change in current, due to the polarization of the ferroelectric layer, is in the order of 10^5 and the off-state current is about 10^-9A (Fig. 2-3). Moreover we are able to program the device at about 10V with a programming time in the order of μs. Ageing tests by accumulating multiple cycling have also been performed. We cycled the Vg potential and measured the Drain/Source current after successive series of cycles on same device. After 10^5 cycles the change in the on-state current is about 10% while the one for the off-state current is about 25% but the current window is large enough that the two states are clearly distinguishable. For the retention measurements we apply a voltage pulse on the gate and measure the time response of the drain and source currents. It is found that the retention time depends on the FET dimensions. This can be as long as 3 days for the large transistors (L=W=50μm) and about 5 to 3 hours for smaller ones (for L=W=10μm and L=W=2μm, respectively). This demonstrates the usability of these devices for any disposable electronic applications requiring memory storage for few days.

2008