PhD Defense: Investigating GeTe-Based Ferroelectric Memristors for Next-Generation AI Hardware

Maxime Culot of SPINTEC will defend his PhD thesis, "Memristive switching in GeTe-based ferroelectric Schottky diodes for neuromorphic computing," on January 21st at 14:00. The defense will be held at the IRIG/SPINTEC auditorium (CEA Grenoble, Building 10.05) and accessible via video conference.

Context: The AI Compute Bottleneck

The research addresses a critical challenge in artificial intelligence: the computing demands of large language models (LLMs) are doubling every two months, straining energy infrastructure, data center cooling, and device portability. While GPUs and TPUs have improved energy efficiency, the thesis explores in-memory computing using emerging non-volatile memories to further reduce the area and power consumption of AI workloads. The memristor crossbar array is a key candidate, offering a compact, low-energy platform for performing matrix-vector multiplication—the core operation in most AI algorithms.

Focus on Ferroelectric GeTe Memristors

Culot's work specifically investigates ferroelectric memristors, which promise high endurance and a strong resistance ratio between states. Germanium Telluride (GeTe) is the material of focus because, unlike many ferroelectrics, its ferroelectric phase can be grown at scale using processes compatible with semiconductor industry back-end-of-line (BEOL) integration.

Experimental Findings and Challenges

The thesis fabricates devices using magnetron-sputtered GeTe to create ferroelectric Schottky diodes (FSDs). Key experimental results include:

* No resistive switching was observed for interfaces with titanium (Ti), platinum (Pt), or palladium (Pd).

* A high pristine resistance and clear resistive switching phenomenon were achieved with a magnesium (Mg) contact. This aligns with theory, as p-type GeTe is expected to form a barrier with low-work-function metals like Mg.

* The observed switching in GeTe/Mg devices depends strongly on the presence of oxygen in the system. While some effects are linked to the metal interface, the thesis could neither confirm nor deny the direct role of ferroelectric polarization in the switching mechanism, highlighting a challenge in controlling polarization within the degenerate semiconductor.

Broader Applications and Simulations

The research extends beyond device fabrication to system-level considerations:

* Sneak Path Mitigation: The inherently parallel structure of crossbar arrays risks altering the state of unselected memristors during write operations. The thesis uses SPICE simulations to model the benefits ferroelectric memristors could provide in suppressing these unwanted "sneak" current paths.

* Boolean Logic Applications: Memristors can implement Boolean logic more compactly than traditional CMOS but are typically hampered by high cycle-to-cycle and device-to-device variability. Culot quantifies the impact of this variability, finding qualitatively that a higher on/off resistance ratio makes logic gates more robust against variations in resistance state or switching voltage.

Conclusion and Jury

Despite the challenges in controlling ferroelectric polarization in GeTe, the thesis concludes that GeTe-based ferroelectric memristors (like FSDs or ferroelectric tunnel junctions) remain highly promising for energy-efficient AI hardware and compact Boolean logic applications.

The defense jury comprises:

* Rapporteurs: Ian O’Connor (École Centrale de Lyon) and Frédéric Leroy (Aix-Marseille Université).

* Examiners: Elisa Vianello (CEA Grenoble) and Jonathan Miquel (Grenoble INP-UGA).

* Thesis Supervisors: Jean-Philippe Attané (Professor) and Louis Hutin.

On Wednesday January 21th, at 14:00, Maxime CULOT (SPINTEC) will defend his PhD thesis entitled :

Memristive switching in GeTe-based ferroelectric Schottky diodes for neuromorphic computing

Place : IRIG/SPINTEC, CEA Building 10.05, auditorium 445 (presential access to the conference room at CEA in Grenoble requires an entry authorization, request it before January 10th to admin.spintec@cea.fr)

video conference : https://univ-grenoble-alpes-fr.zoom.us/j/3241920232

Abstract : The computing requirements of large language models (LLMs) is doubling every 2 months. This exponential growth raises challenges in energy production, heat dissipation in data centres and portability of artificial intelligence (AI) devices. The development of new inherently parallel computing architectures such as graphics (GPUs) and tensor (TPUs) processing units to perform AI computations have allowed the increase of the energy efficiency of AI workloads. New in memory computing approaches based on emerging non-volatile memories have been proposed to further reduce the area requirements and energy consumption linked with the computation of AI workloads. Among them, the memristor crossbar array holds the promise of a very high compactness and low energy implementation of matrix vector multiplication, the operation at the heart of most AI algorithms. The memristor concept, a two-terminal circuit element with a variable resistance, can be implemented with different phenomena: phase change, conductive filament formation and breaking, magnetic or ferroelectric switching, etc. This thesis focuses on ferroelectric memristors because of their potential for high endurance and high resistance ratio between their high and low resistance states. While most ferroelectric materials are tricky to integrate with the processes of the semiconductor industry, the large-scale, back-end-of-line (BEOL) compatible growth of the ferroelectric phase of GeTe has been demonstrated. This thesis aims to demonstrate the potential of GeTe based memristors for neu- romorphic computing. First, magnetron-sputtered GeTe is used to fabricate devices exploiting the GeTe/metal interface for the realisation of ferroelectric Schottky diodes (FSDs). No resistive switching was observed for the GeTe/Ti, GeTe/Pt or GeTe/Pd contact, but a high resistance in the pristine state and a resistive switching phenomenon was observed for the GeTe/Mg contact. Despite its specificities as a degenerate semi- conductor and its strong Fermi level pinning, being p-type, the formation of a barrier with GeTe is expected for metals with a low work function, like Mg. Actually, GeTe/Mg devices present several resistive switching phenomena which are shown to strongly depend on the presence of oxygen in the system. While the origin of some of those phenomena is linked with the GeTe/metal interface, the role of ferroelectric polarisation in the observed resistive switching can be neither confirmed nor denied. The integration of memristors in crossbar arrays is very compact. However, this structure is inherently parallel, and the application of a voltage difference across a device results in the application of this voltage on all parallel paths. Without a specific scheme to suppress the unwanted current paths, there is a high risk of changing the state of unselected memristors during the writing step. The benefits that ferroelectric memristors bring for the suppression of sneak paths in memristor crossbar arrays are considered and simulated with SPICE simulations. Finally, memristor crossbar arrays have other applications besides neuromorphic computing. Memristors have been shown to realise Boolean logic in a more compact way than traditional CMOS logic. However, they usually suffer from cycle-to-cycle and device- to-device variations that are too high for the exactness requirements of Boolean logic. The impact of variability of the memristor on the good operation of memristive Boolean logic gates is quantified. Qualitatively, as expected, a higher on/off ratio memristor results in a logic gate that is less susceptible to memristor resistance state or switching voltage variations. Despite the challenges associated with the control of the ferroelectric polarisation in the degenerate semiconductor GeTe, the realisation of ferroelectric based memristors like FSDs or ferroelectric tunnel junctions is very promising for AI workloads, a very hot topic, but also for Boolean applications.

titre : Comportement memristif de diodes Schottky ferroélectriques à base de GeTe pour le calcul neuromorphique

Résumé : Les besoins de calculs des grands modèles de langage (LLM) doublent tous les deux mois. Cette croissance exponentielle pose des défis en matière de production d’énergie, de dissipation thermique dans les data centres etde portabilité des appareils d’intelligence artificielle (IA). Le développement de nouvelles architectures de calcul intrinsèquement parallèles, telles que les GPUs et les TPUs, a permis d’augmenter l’efficacité énergétique deces applications. De nouvelles approches de calcul en mémoire basées sur des mémoires non volatiles émergentes ont été proposées afin de réduire davantage les besoins en surface et la consommation d’énergie liés auxcalculs d’IA. Parmi celles-ci, la matrice de memristor promet une mise en oeuvre compacte et à faible consommation d’énergie de la multiplication matrice vecteur, au centre de la plupart des algorithmes d’IA. Le concept dememristor, un élément de circuit à deux bornes avec une résistance variable, peut être mis en oeuvre à l’aide de différents phénomènes : changement de phase, formation et rupture de filaments conducteurs, commutationmagnétique ou ferroélectrique, etc. Cette thèse se concentre sur les memristors ferroélectriques en raison de leur potentiels haute endurance et rapport de résistance élevé entre leurs états de résistance élevée et faible.Alors que la plupart des matériaux ferroélectriques sont difficiles à intégrer dans les processus de l’industrie des semi-conducteurs, la croissance à grande échelle et compatible avec le back-end-of-line (BEOL) de la phaseferroélectrique du GeTe a été démontrée. Cette thèse vise à démontrer le potentiel des memristors à base de GeTe pour le calcul neuromorphique. Du GeTe déposé par pulvérisation cathodique magnétron est utilisé pourfabriquer des dispositifs exploitant l’interface GeTe/métal pour la réalisation de diodes Schottky ferroélectriques (FSD). Aucun changement de résistance n’a été observé pour les contacts GeTe/Ti, GeTe/Pt ou GeTe/Pd, maisune résistance élevée à l’état vierge et un phénomène de commutation résistive ont été observés pour le contact GeTe/Mg. Étant de type p, la formation d’une barrière avec le GeTe est attendue pour les métaux à faibletravail de sortie, comme Mg. Si l’origine de certains de ces phénomènes est liée à l’interface GeTe/métal, elle dépend fortement de la présence d’oxygène dans le système et le rôle de la polarisation ferroélectrique ne peutêtre ni confirmé ni infirmé. L’intégration des memristors dans les matrices crossbar est très compacte. Cependant, cette structure est intrinsèquement parallèle, et l’application d’une différence de potentiel aux bornes d’undispositif entraîne l’application de cette tension sur tous les chemins parallèles. Sans une stratégie pour supprimer les chemins parasites, une modification de l’état des memristors non sélectionnés pendant l’étape d’écritureest possible. Les avantages des memristors ferroélectriques pour la suppression des chemins parasites dans les matrices crossbar de memristors sont examinés et simulés à l’aide de simulations SPICE. Enfin, les matrices dememristors ont d’autres applications que l’IA et permettent aussi de réaliser la logique booléenne de manière plus compacte que les circuits CMOS traditionnels. Cependant, les variations d’un cycle à l’autre et d’un dispositifà l’autre sont souvent trop importantes pour répondre aux exigences de précision de la logique booléenne. L’impact de la variabilité du memristor sur le la logique booléenne memristive est quantifié. Qualitativement, unmemristor avec un rapport ON/OFF plus élevé est moins sensible aux variations de l’état de résistance du memristor ou de la tension de commutation. Malgré les défis liés au contrôle de la polarisation ferroélectrique dans lesemi-conducteur dégénéré GeTe, la réalisation de memristors ferroélectriques tels que les FSDs ou les jonctions tunnel ferroélectriques est très prometteuse pour les calculs d’IA, mais aussi pour les applications booléennes.

Jury :

Ian O’Connor, Ecole Centrale De Lyon, Rapporteur

Frédéric Leroy, Aix-Marseille Universite, Rapporteur

Elisa Vianello, Cea Grenoble, Examinatrice

Jonathan Miquel, Grenoble Inp-Uga, Examinateur

Thesis Supervisors :

Jean-Philippe Attane, Professeur

Louis Hutin

The post PhD Defense by Maxime CULOT – Memristive switching in GeTe-based ferroelectric Schottky diodes for neuromorphic computing appeared first on Spintec.