Tyler Hennen's Publications


Synaptogen: A Cross-Domain Generative Device Model for Large-Scale Neuromorphic Circuit Design

Tyler Hennen, Leon Brackmann, Tobias Ziegler, Sebastian Siegel, Stephan Menzel, Rainer Waser, Dirk J. Wouters, Daniel Bedau

IEEE Transactions on Electron Devices

(Open Access)

July 2024

We built a fast, realistic model of advanced memory devices that helps engineers simulate large-scale circuits more efficiently than ever before.

A Digital Twin for Next-Generation Hardware

Designing the next generation of computing hardware means working with millions of unpredictable memory devices. To make this possible, we created Synaptogen, a 'digital twin' that realistically models the complex behavior of these components. Our ultra-fast software allows engineers to simulate massive circuits with confidence, speeding up the design process and reducing the risk of costly errors during fabrication.

We created a new generative model that reproduces how real devices behave, including their quirks and randomness. We trained it on a massive dataset of 6,000 switching cycles from 512 actual devices, capturing how each one changes over time and how they differ from each other. Then we implemented the model in a hardware-compatible format (Verilog-A), so it can be used directly in circuit simulators.

Why does this matter?
Traditional models are either too simplistic or too slow. Synaptogen is fast and statistically realistic—over 100× faster than previous detailed models—making it possible to simulate large-scale systems. This helps engineers test and build more powerful and energy-efficient hardware.

Full Factorial Analysis of Gradual Switching in Thermally Oxidized Memristive Devices

Pascal Stasner, Tyler Hennen, Ekaterina Gorbunova, Alejandro García Munoz, Rainer Waser, Dirk J. Wouters

Journal of Applied Physics

(Open Access)

June 2024

We figured out how to precisely grow the oxide material in next-generation memory devices to make them switch more gradually and reliably. To do this, we developed a high-speed testing method that analyzed millions of switching cycles, helping us pinpoint the exact conditions needed for analog behavior.

Optimizing Analog Memory

For computers to process complex data more efficiently, they need components that can change their resistance in a smooth, analog way, not just flipping between on and off. A key challenge with these components, called memristors, is that they tend to switch too abruptly and unreliably. We tackled this by systematically studying how different fabrication conditions affect device performance. We developed a powerful 'full factorial' testing method that let us rapidly test 2,500 unique control settings in a single ten-second measurement, gathering a massive dataset of over a million switching events per device type.

Our analysis revealed that creating the oxide at a lower temperature (250°C vs. 300°C) produced a device with a slightly higher internal series resistance. This small change, combined with a milder electrical characteristic, was the key to preventing the abrupt, uncontrolled switching common in these devices.

Why does this matter?

This work provides a clear way to build more reliable memristors for analog computing. By fine-tuning the bottom-up growth of the oxide, we can create devices that switch gradually and predictably, which is a critical step toward building powerful and efficient computing systems. Furthermore, our high-speed factorial analysis method gives researchers a powerful new tool to quickly and thoroughly evaluate and optimize other new materials and device designs, accelerating progress in the field.

Bulk-Like Mott-Transition in Ultrathin Cr-Doped V2O3 Films and the Influence of Its Variability on Scaled Devices

Johannes Mohr, Tyler Hennen, Daniel Bedau, Rainer Waser, Dirk J. Wouters

Advanced Physics Research

(Open Access)

May 2024

We performed the first-ever direct electrical measurements of a pressure-driven phase transition in ultra-thin films of a promising electronic material. We showed that even at a thickness of just 10 nanometers, the material behaves remarkably like its ideal, perfect-crystal form, which is great news for building tiny, reliable nanoelectronic devices.

A Robust Switch, Even When Thin

Certain advanced materials, known as Mott insulators, have the unique ability to switch from an electrical insulator to a metal when pressure is applied, making them exciting candidates for new types of electronic devices. A major question, however, was whether this useful property would survive when the material was made into an ultra-thin film, as required for all modern electronics. We answered this by performing the first direct electrical measurements of this transition in films of chromium-doped V₂O₃ as thin as 10 nanometers, using a special high-pressure chamber.

Our measurements revealed a dramatic change in resistance of nearly 100 times, proving that the pressure-driven switching effect is incredibly robust, even in these imperfect, polycrystalline films. To understand this better, we developed a simulation that modeled the film as a network of tiny crystal grains, each with a slightly different transition pressure. This model showed that the behavior we measured is perfectly consistent with the ideal bulk material, just averaged across the device.

Why does this matter?

This work confirms that the fundamental switching property of this material is not lost in the ultra-thin, imperfect films that are used in real-world applications. This is a critical finding because it confirms the material can be used to make devices. Our simulations further show that despite the material's inherent grain-to-grain variability, devices can be reliably scaled down to at least 50 nanometers in size without suffering from excessive performance variations. This gives us confidence in using this promising material for future nanoelectronic systems.

Reliability Effects of Lateral Filament Confinement by Nano-Scaling the Oxide in Memristive Devices

Pascal Stasner, Nils Kopperberg, Kristoffer Schnieders, Tyler Hennen, Stefan Wiefels, Stephan Menzel, Rainer Waser, Dirk J. Wouters

Nanoscale Horizons

(Open Access)

March 2024

We designed a new type of memory device where we build tiny 10-nanometer 'walls' around the conductive filament to keep it from changing shape. This physical confinement makes the device significantly more stable and reliable, which is a critical step for building next-generation, high-density memory.

Putting Walls Around Memory

One of the biggest hurdles for using memristors in new computing hardware is their reliability. The tiny conductive filaments that store information inside these devices are prone to random fluctuations, causing performance issues. The conventional way to shrink these devices is to simply make the electrodes smaller, but this doesn't stop the filament itself from changing its shape or position. To solve this, we took a new approach: instead of just scaling the electrodes, we scaled the active oxide material itself.

We developed a 'nano-fin' fabrication process that physically confines the switching oxide to a 10 nm wide strip, effectively building walls around the filament and limiting its ability to spread out. We then ran a massive electrical analysis, stressing the devices with 100,000 specially modulated cycles to test their performance under 2,500 unique conditions.

Why does this matter?

By physically controlling the filament's geometry, we directly address a root cause of unreliability. Our measurements show that this confinement leads to a 50% decrease in the variability when writing to the device and a significant improvement in the signal-to-noise ratio, making the device more stable and predictable. This concept of 'oxide scaling'—controlling the filament environment itself—is a major step beyond simple 'electrode scaling' and shows a powerful new direction for engineering the robust and reliable hardware needed for future computing technologies.

Physical Origin of Threshold Switching in Amorphous Chromium-Doped V2O3

Johannes Mohr, Christopher Bengel, Tyler Hennen, Daniel Bedau, Stephan Menzel, Rainer Waser, Dirk J. Wouters

physica status solidi (a)

(Open Access)

December 2023

We discovered that the electrical switching observed in devices made from amorphous (non-crystalline) materials is actually caused by a two-step process. An initial high-voltage pulse first creates a tiny crystalline filament within the material, and then a well-understood thermal effect within that newly formed filament causes the actual switching.

Uncovering the Switching Mechanism

To be compatible with modern chip manufacturing, it's highly desirable to use electronic materials that can be deposited at room temperature. This process, however, usually creates an amorphous or non-crystalline structure. While devices made from amorphous chromium-doped V₂O₃ showed promising electrical switching behavior, the physical reason behind it was a mystery. We investigated this by fabricating and testing devices with different material compositions.

Our experiments revealed that an initial, one-time high-voltage pulse, known as a 'forming step,' is required to activate the switching. We demonstrated that this step causes a permanent physical change: the intense local heat from the current crystallizes a a small, conductive filament within the otherwise amorphous material. The repeatable on-off switching observed afterward is not a property of the amorphous bulk, but is actually a 'thermal runaway' effect—a rapid self-heating process—that occurs only within this newly formed crystalline filament.

Why does this matter?

This work clarifies why the switching happens in these promising devices, showing that we are creating a crystal filament 'in-place'. This understanding is crucial for making better devices. Using a model that accurately describes the thermal process, we were able to show exactly how changing the chromium and oxygen content affects the switching voltage and leakage current. This provides a clear guide for optimizing the material composition to create devices with the ideal performance for practical applications, like selector elements in next-generation memory arrays.

Harnessing Stochasticity and Negative Differential Resistance for Unconventional Computation

Tyler Hennen

RWTH Aachen University

(Open Access)

July 2023

Tyler explored futuristic materials for new types of energy-efficient computers. He made tools to test them, created software to simulate them, and figured out how to make them useful for next-gen computing.

Tyler's thesis explores new types of tiny electronic devices that could one day help computers work more like the human brain—faster, more energy-efficient, and better at handling complex tasks like learning and pattern recognition. He focuses on two kinds of experimental devices: one that mimics synapses and one that mimics neurons.

ReRAM: Artificial Synapses

These memory devices change resistance when voltage is applied, similar to how synapses strengthen or weaken in the brain. Tyler created a new model to simulate their real-world behavior using massive amounts of data. His simulator can handle billions of memory cells and run extremely fast.

Cr-doped V₂O₃: Artificial Neurons

This material switches rapidly between insulating and conducting states, similar to a neuron firing. Tyler studied how heat and electricity cause this behavior, and built accurate models to predict it. He showed the devices can switch in under 10 nanoseconds and last trillions of cycles.

Tools and Innovations

He built custom test equipment that works much faster than commercial tools, and developed software to simulate entire networks of these devices at scale.

Why does this matter?

Today’s computers waste energy moving data between memory and processors. Brain-inspired hardware—called neuromorphic computing—could solve this by computing and storing in the same place. Tyler’s work helps us understand and simulate the devices that make this possible.

Eliminating Capacitive Sneak Paths in Associative Capacitive Networks Based on Complementary Resistive Switches for In-Memory Computing

Tobias Ziegler, Leon Brackmann, Tyler Hennen, Christopher Bengel, Stephan Menzel, Dirk J. Wouters

IEEE International Memory Workshop (IMW)

May 2023

We identified a flaw in a new type of computing-in-memory design that caused errors during read-out. We then proposed a straightforward fix to how data is fed into the system, which completely solves the problem and improves performance.

Fixing the Sneak Paths

A new design for in-memory computing, called an Associative Capacitive Network (ACN), shows real promise for low-power applications. It works by storing information in special switchable capacitors. When we simulated this design, however, we discovered a key problem. Unwanted electrical signals were 'sneaking' through unintended paths in the circuit, which scrambled the results and made it impossible to reliably tell what data was stored.

The issue stemmed from the original method for encoding input data, which left some circuit lines floating in a high-impedance state. Our solution is to actively drive all lines, connecting them to either the read voltage or to ground.

Why does this matter?

Our work found a flaw in the original design that prevented it from working. More importantly, we developed a new input method that completely gets rid of these sneak paths. This fix not only prevents errors but also makes the output signals cleaner and easier to distinguish. This change makes the design more practical and reliable for future energy-efficient computing systems.

Fabrication of Highly Resistive NiO Thin Films for Nanoelectronic Applications

Johannes Mohr, Tyler Hennen, Daniel Bedau, Joyeeta Nag, Rainer Waser, Dirk J. Wouters

Advanced Physics Research

(Open Access)

October 2022

Nickel oxide is a promising but tricky material for new electronic devices; it's very difficult to produce in its most useful, highly insulating form. We used a systematic statistical approach to map out the effects of different fabrication settings, allowing us to identify a clear and repeatable recipe for making high-quality films.

A Recipe for a Challenging Material

Nickel oxide (NiO) has the potential to be a key ingredient in a variety of future nanoelectronic devices, from new types of computer memory to circuits for novel computing hardware. However, it's notoriously difficult to work with. Depending on how it's made, its electrical resistance can vary by billions of times, and researchers often struggle to get the highly insulating version that is most desirable. To solve this, we took a more systematic approach.

We used a statistical method called 'design of experiments' to test a wide range of sputtering parameters—like temperature, pressure, and gas mixtures. We then used a machine learning technique called cluster analysis to sort the resulting films into four distinct groups based on their structural and electrical properties. This allowed us to clearly see which settings produced which type of film and, most importantly, to identify the optimal recipe for the one we wanted.

Why does this matter?

This work provides a clear and reliable recipe for making high-quality insulating NiO thin films. We found that the key is to use a high deposition temperature to create a dense film, an oxygen-rich atmosphere to make it smooth, and a final annealing step to remove excess oxygen and achieve high resistance. By identifying the most critical parameters and their interactions, we have removed the guesswork from the process. This enables researchers to reliably produce the ideal material needed to build and test the next generation of advanced electronic devices.

A High Throughput Generative Vector Autoregression Model for Stochastic Synapses

Tyler Hennen, Alexander Elias, Jean‑François Nodin, Gabriel Molas, Rainer Waser, Dirk J. Wouters, Daniel Bedau

Frontiers in Neuroscience

(Open Access)

August 2022

We built an ultra-fast software model that realistically simulates the tiny, unpredictable electronic components used in next-generation computers for AI. This tool allows designers to accurately and rapidly test massive networks of these components before committing to costly hardware fabrication.

A Digital Twin for Neuromorphic Chips

The next wave of artificial intelligence hardware uses new electronic devices that act like synapses, but a major challenge is that these devices are unpredictable and have complex statistical behaviors, which makes designing large, reliable circuits incredibly difficult. Before building a chip with millions or billions of these 'synapses', you need to simulate them, but previous methods were either too slow to be practical or too simple to be accurate.

We tackled this problem by taking a new, data-driven approach. We recorded a massive amount of electrical data from a real-world device, capturing its unique statistical personality over millions of operating cycles. We then used this data to create a lightweight and powerful generative model that can accurately reproduce the device's complex behavior, including its randomness and how its properties correlate over time. We developed highly-optimized versions of this model that run on both standard processors (CPUs) and graphics cards (GPUs), making it exceptionally fast.

Why does this matter?
This work provides a critical tool. It bridges the gap between a single lab device and a future computer chip containing billions of them. By allowing researchers to simulate large-scale systems that behave like the real hardware, our model speeds up the design and testing of powerful, energy-efficient computer chips for the future of AI. It gives designers the confidence to explore new ideas on a computer, saving time and money during chip development.

Counting Molecules: Python Based Scheme for Automated Enumeration and Categorization of Molecules in Scanning Tunneling Microscopy Images

Jack Hellerstedt, Aleš Cahlík, Martin Švec, Oleksandr Stetsovych, Tyler Hennen

Software Impacts

(Open Access)

May 2022

We developed a free, open-source software tool that automatically counts and sorts different types of molecules in high-resolution microscope images. This saves researchers a lot of time and allows for more complex data analysis than what was possible by hand.

Automating Molecular Analysis

Scientists use powerful tools like scanning tunneling microscopes (STM) to take pictures of chemical reactions at the molecular level. A single image can contain hundreds of molecules of various types, but researchers have traditionally had to count and categorize them by hand. This manual process is slow, tedious, and limits the complexity of the data that can be analyzed.

To solve this, we created a smart, automated tool using Python. Our software analyzes STM images, identifies each molecule, and then sorts them into different categories based on their unique features, like shape and size. It’s designed to be easy to use right out of the box, even for those who aren't programming experts.

Why does this matter?

This work makes studying chemical reactions at the molecular level faster and more powerful. By automating the counting process, our tool frees up scientists to tackle bigger datasets and explore more subtle statistical trends. This leads to new discoveries in on-surface chemistry and materials science. Because the software is open-source, the scientific community can freely use, modify, and improve it, accelerating research for everyone.

Stabilizing Amplifier with a Programmable Load Line for Characterization of Nanodevices with Negative Differential Resistance

Tyler Hennen, Erik Wichmann, Rainer Waser, Dirk J. Wouters, Daniel Bedau

Review of Scientific Instruments

arXiv version

February 2022

We designed a programmable circuit that makes it easier to test advanced nanoelectronic devices, helping researchers study how these tiny systems behave under real-world conditions.

Testing next-generation memory and logic devices is tricky—especially those that are unstable or unpredictable.

This paper introduces a custom-built measurement circuit that helps researchers explore these challenging behaviors more safely and accurately. It adds a programmable resistance into the test setup, giving precise control over the electrical environment around a device.

Why is this needed? Devices like ReRAM and other nanoscale components often show something called negative differential resistance—a kind of behavior where current can decrease even as voltage increases. This can lead to runaway effects, damaging the device or corrupting the data. Our solution uses a digitally controlled resistor (called a digipot) that can switch between 528 resistance levels and capture fast electrical signals without distorting them.

Why does this matter?
Safely and accurately measuring unstable devices is crucial for building reliable new computing technologies. Our circuit gives scientists a flexible, computer-controlled way to test a wide variety of devices—including those used in advanced computing—while minimizing unwanted noise and measurement errors. It also helps explore how different test conditions affect device performance, making it easier to design smarter electronics from the ground up.

Lattice Contraction Induced by Resistive Switching in Chromium-Doped V2O3: A Hallmark of Mott Physics

Danylo Babich, Julien Tranchant, Coline Adda, Benoit Corraze, Marie‑Paule Besland, Peter Warnicke, Daniel Bedau, Patricia Bertoncini, Jean‑Yves Mevellec, Bernard Humbert, Jonathan A. J. Rupp, Tyler Hennen, Dirk J. Wouters, Roger Llopis, Laurent Cario, Etienne Janod

arXiv

October 2021

We used high-powered x-rays to look inside a special electronic material as it switched from an insulator to a metal. We found that the conductive path it creates is not chemically different but is physically squeezed, which is direct evidence that the switching is caused by a true Mott transition.

Squeezing Atoms with Electricity

Mott insulators are a special class of materials that can be flipped from an insulating to a conducting state with an electric field, which makes them very interesting for new kinds of electronics. However, the exact reason for this useful switching behavior has been heavily debated. Was it a chemical change, like oxygen atoms moving around, or was it a more fundamental electronic effect tied to the material itself?

To answer this, we used a combination of powerful experimental techniques, including micro x-ray diffraction and Raman spectroscopy, to look directly at the atomic structure of the material as it switched. This allowed us to map the crystal lattice of the tiny conductive filament that forms when an electric pulse is applied.

Why does this matter?

Our experiments revealed that the filament is structurally compressed, with a smaller unit cell volume than the surrounding insulating material. This is the exact same structural 'fingerprint' that appears when this material is turned into a metal by applying immense physical pressure. This is the first direct proof that the electrical switching is caused by a true Mott transition, not a chemical reaction. This fundamental insight clarifies how these devices work and makes them more promising for use in a new generation of advanced electronics.

Current-Limiting Amplifier for High Speed Measurement of Resistive Switching Data

Tyler Hennen, Erik Wichmann, Alexander Elias, Jeffrey Lille, Oleksandr Mosendz, Rainer Waser, Dirk J. Wouters, Daniel Bedau

Review of Scientific Instruments

arXiv version

May 2021

We designed and built a specialized circuit that protects delicate electronic components from destroying themselves during testing. This allows us to test these components millions of times faster than before, which is crucial for developing next-generation computer memory and logic chips.

Taming the Runaway Switch

A new class of electronic components, called resistive switches, are key for future technologies like advanced computer memory (RRAM) and other novel computing architectures. A major problem is that these tiny switches are very delicate. When they turn on, they can experience a sudden, uncontrolled surge of current that instantly destroys them. The standard solution is to build a protective transistor right next to each switch on the chip, but this is complicated and slows down research and development.

We created an external circuit, a Current Limiting Amplifier (CLA), that solves this problem without needing on-chip modifications. It sits between the testing equipment and the device, allowing the switch to operate normally but instantly clamping down on any destructive current surges. Our design is fast, stable, and minimizes electrical noise, preventing the 'overshoots' that plague commercial testing equipment and can damage the devices.

Why does this matter?

This new amplifier allows us to test resistive switches at incredible speeds—we demonstrated collecting 100,000 full switching cycles in just one second. That's about a million times faster than standard lab equipment. This high speed makes it practical to gather the massive amounts of statistical data needed to understand the unreliable nature of these devices and figure out how to make them better. By enabling rapid and safe testing, our work accelerates the development of the hardware needed for the next generation of computing.

A Mott Insulator-Based Oscillator Circuit for Reservoir Computing

Wen Ma, Tyler Hennen, Martin Lueker‑Boden, Rick Galbraith, Jonas Goode, Won Ho Choi, Pi‑Feng Chiu, Jonathan A. J. Rupp, Dirk J. Wouters, Rainer Waser, Daniel Bedau

IEEE International Symposium on Circuits and Systems (ISCAS)

October 2020

We built a novel computing system using a special electronic device that naturally oscillates. This system performs complex tasks like speech and handwriting recognition with high accuracy, but with a much smaller model and significantly lower power consumption than conventional hardware.

Teaching Tiny Oscillators to Think

Modern artificial intelligence is incredibly powerful, but training and running AI models can consume a tremendous amount of energy. To tackle this, we are exploring new types of computer hardware that take inspiration from the efficiency of biological computation. We focused on a concept called 'reservoir computing,' which is a type of neural network that is very fast to train.

Our approach uses a unique electronic device based on a material called a Mott insulator (specifically, Cr-doped V₂O₃). When placed in a simple circuit, this device naturally starts to oscillate, producing electrical spikes that closely resemble the 'leaky-integrate-and-fire' behavior of biological neurons. We used an array of these simple, neuron-like circuits to form the 'reservoir'—the dynamic and complex core of our computing system.

Why does this matter?

Using these physical neuron-like circuits, we created a system that can learn and perform complex tasks with remarkable efficiency. We successfully demonstrated its ability to recognize spoken words and handwritten digits with an accuracy that rivals much larger, more complex AI models like LSTMs. Crucially, our system achieves this with a tiny fraction of the power and a significantly faster processing speed compared to running the same tasks on a CPU, GPU, or FPGA. This work shows a promising path toward building powerful, low-energy hardware for fast AI applications.

Comprehensive Model for the Electronic Transport in Pt/SrTiO3 Analog Memristive Devices

Carsten Funck, Christoph Bäumer, Stefan Wiefels, Tyler Hennen, Rainer Waser, Susanne Hoffmann‑Eifert, Regina Dittmann, Stephan Menzel

Physical Review B

July 2020

We figured out how a certain type of next-gen memory device stores multiple resistance levels. The secret is a tiny energy barrier at the metal contact that gets wider or narrower to control how easily electrons can pass through.

How Electrons Cross the Barrier

To build computers that can process information more efficiently, we need memory components that can store a whole range of values, not just a simple '0' or '1'. We looked closely at a promising device made of Strontium Titanate (SrTiO₃) to understand how it achieves this analog behavior. We ran a series of experiments, measuring the device's electrical properties at different temperatures and for various resistance states.

To explain what we saw, we built a detailed computer simulation. The model showed that the device's resistance is controlled by a thin energy barrier that forms right where the oxide meets the metal electrode. The device changes its resistance by moving oxygen atoms around, which in turn makes this barrier wider or narrower. For electrons to get across, they need enough thermal energy to get close and then 'tunnel' through the rest of the way.

Why does this matter?

This work explains the physics of how these analog memory devices operate. By understanding that everything comes down to tuning this tunnel barrier, engineers now have a clear target for improvement. It gives them the knowledge to design more precise, stable, and reliable analog memory, which is a key component for building future advanced computing architectures.

Study of the SET Switching Event of VCM-based Memories on a Picosecond Timescale

Moritz von Witzleben, Tyler Hennen, Andreas Kindsmüller, Stephan Menzel, Rainer Waser, Ulrich Böttger

Journal of Applied Physics

May 2020

We developed a new measurement technique to find out just how fast next-generation memory devices can switch. Using this method, we were able to clock switching speeds down to 50 picoseconds—that's the time it takes light to travel less than an inch.

Clocking the Speed of Memory

A key question for new memory technologies is: what is their ultimate speed limit? Existing methods struggle to measure switching events that happen faster than a nanosecond. We developed a new experimental setup that uses a specialized pulse generator and advanced statistical analysis to precisely determine switching times on a picosecond timescale (trillionths of a second). We used this method to compare the top speeds of two promising memory materials, Tantalum Oxide (TaOₓ) and Zirconium Oxide (ZrOₓ).

By applying thousands of extremely short electrical pulses and measuring the result of each one, we could pinpoint exactly how long it takes for a device to flip from a high to a low resistance state.

Why does this matter?

Our technique allowed us to measure memory switching in as little as 50 picoseconds, with the actual resistance transition taking less than 15 picoseconds—the fastest ever reported for these kinds of devices. We also found that ZrOₓ switches faster than TaOₓ at the same voltage, which is likely due to the lower energy needed to move oxygen atoms in that material. This work provides an important tool for discovering and comparing the true speed of new memory materials, which helps direct research toward the fastest possible technologies.

Switching Speed Analysis and Controlled Oscillatory Behavior of a Cr-Doped V2O3 Threshold Switching Device for Memory Selector and Neuromorphic Computing Application

Tyler Hennen, Daniel Bedau, Jonathan A. J. Rupp, Carsten Funck, Stephan Menzel, Michael K. Grobis, Rainer Waser, Dirk J. Wouters

IEEE International Memory Workshop (IMW)

May 2019

We studied what makes a specific type of electronic switch operate quickly. We found that driving it with a higher voltage through a larger resistor makes it switch faster and more efficiently, and that adding a capacitor can make it oscillate like a neuron.

Making a Switch Faster

Electronic switches used in new computer memories need to be both fast and efficient. We investigated a promising switch made from chromium-doped vanadium oxide (Cr-doped V₂O₃) to understand what controls its speed. Using a combination of experiments and a thermal model, we analyzed how the device responds to different voltage pulses and resistive loads. We also explored what happens when you connect a capacitor to it.

We found a surprising result: you can make the device switch faster and use less total energy by using a larger series resistor and a higher voltage. This works by pushing the device far from its stable thermal state, which speeds up the underlying self-heating process. We also demonstrated that adding a small capacitor to the circuit turns the switch into a tiny, voltage-controlled oscillator that operates in the megahertz range.

Why does this matter?

This provides clear design rules for engineers looking to use these switches in real-world systems. Our findings offer a direct strategy for optimizing speed and energy efficiency in memory selector applications. Additionally, showing that the device can act as a compact, controllable oscillator confirms it can be used as an artificial neuron in new types of brain-inspired computer circuits.

Forming-Free Mott-Oxide Threshold Selector Nanodevice Showing S-Type NDR with High Endurance (> 1012 Cycles), Excellent Vth Stability (< 5%), Fast (< 10 ns) Switching, and Promising Scaling Properties

Tyler Hennen, Daniel Bedau, Jonathan A. J. Rupp, Carsten Funck, Stephan Menzel, Michael K. Grobis, Rainer Waser, Dirk J. Wouters

IEEE International Electron Devices Meeting (IEDM)

December 2018

We built the smallest-ever selector devices from a special Mott-oxide material. They work right out of the box without needing an activation step, are extremely fast and durable, and our models show they will perform well even when made much smaller.

A Near-Perfect Selector Switch

To build denser computer memory, every memory cell needs its own 'selector' switch to make sure only the right cell is being read or written. We fabricated and tested the smallest-ever selector devices based on crystalline chromium-doped vanadium oxide, with features down to 120 nanometers across. These devices performed exceptionally well.

They are 'forming-free,' which means they work immediately without a high-voltage 'break-in' procedure. They switch on and off in under 10 nanoseconds and show remarkable stability, with key parameters like the threshold voltage changing by less than 5% over a trillion cycles. We also found that their switching is based on a uniform heating effect across the whole device volume, not an unstable, filamentary one.

Why does this matter?

These devices successfully meet many of the key requirements for a high-performance memory selector. Their stability and high endurance are especially important. We developed a thermal model that perfectly explains how their performance depends on their size, which allows us to predict how they will behave at even smaller, more technologically advanced dimensions. This work presents a successful and scalable selector technology for future dense and reliable computer memory.

On the Universality of the I – V Switching Characteristics in Non-Volatile and Volatile Resistive Switching Oxides

Dirk J. Wouters, Stephan Menzel, Jonathan A. J. Rupp, Tyler Hennen, Rainer Waser

Faraday Discussions

June 2018

We explain that even though the electrical switching curves of different memory devices can look identical, it doesn't mean they work the same way inside. This work cautions researchers to be careful when interpreting these measurements to understand what's really going on.

Looks Can Be Deceiving

When studying new memory devices, researchers often use current-voltage (I-V) curves to see how they switch. A certain feature—a constant voltage during the 'ON' switch—is often seen in non-volatile memory and is typically explained by a conductive filament inside the device growing wider. The issue is, we've observed this exact same feature in a completely different, volatile type of switch (Cr-doped V₂O₃). This raises an important question: do they really work the same way just because their I-V curves look alike?

This paper reviews the common models and highlights this potential pitfall. We argue that very different physical mechanisms can actually produce very similar-looking electrical signatures. For instance, the behavior in the volatile switch might be better explained by a thermal effect within a filament of a *fixed* size, rather than one that changes its width.

Why does this matter?

This work challenges a common assumption in the field of resistive memory. It shows that relying only on I-V curves to identify a device's internal switching mechanism can be misleading. To truly understand and improve these complex devices, it's important to use better measurement techniques to get the full picture. Understanding the physics is a necessary step to correctly model and ultimately optimize these devices for future technologies.

Bit Patterned Media Optimization at 1 Tdot/In2 by Post-Annealing

Olav Hellwig, Ernesto E. Marinero, Dan Kercher, Tyler Hennen, Andrew McCallum, Elizabeth Dobisz, Tsai‑Wei Wu, Jeffrey Lille, Toshiki Hirano, Ricardo Ruiz, Michael K. Grobis, Dieter Weller, Thomas R. Albrecht

Journal of Applied Physics

September 2014

We discovered that baking a patterned hard drive disk in a vacuum after it's been made not only heals damage from the manufacturing process but also chemically reorganizes the atoms to create much more stable magnetic islands, enabling ultra-high-density data storage.

Better Magnets Through Baking

One of the most promising ways to create next-generation hard drives is 'bit-patterned media,' where data is stored on billions of tiny, physically separated magnetic islands. A major challenge is that the process used to etch these islands can damage their edges, which hurts performance and reliability. We investigated a simple but powerful solution: what if we just bake the disk in a vacuum after the islands are made?

We took a disk with a density of one trillion dots per square inch (1 Tdot/in²) and annealed it at 400°C. The results were dramatic. The coercivity (a measure of magnetic robustness) more than doubled, and the thermal stability—the ability to resist being erased by heat—increased by a massive 3-fold.

Why does this matter?

This post-annealing process does more than just fix damage. Our analysis revealed it starts a chemical reaction where the chromium atoms in the alloy migrate from the center of each island to the edges. This creates an ideal structure: a highly stable, high-performance magnetic core surrounded by a protective, non-magnetic shell. This discovery provides a simple and effective way to create the top-quality, ultra-dense media required for the hard drives of the future.

Influence of Stray Fields on the Switching-Field Distribution for Bit-Patterned Media Based on Pre-Patterned Substrates

Bastian Pfau, Christian M. Günther, Erik Guehrs, Thomas Hauet, Tyler Hennen, Stefan Eisebitt, Olav Hellwig

Applied Physics Letters

September 2014

When writing to a single magnetic island on a patterned hard drive, stray magnetic fields from the surroundings can cause errors. We used a special x-ray imaging technique to discover that the magnetic material left in the trenches *between* the islands is a major, and often overlooked, source of this interference.

Mind the Trenches

For bit-patterned media to work, each magnetic island must flip at a very precise, predictable magnetic field. Stray fields from the island's environment can alter this switching point, causing errors. While it's known that neighboring islands interfere with each other, we wanted to understand the impact of the magnetic material left in the trenches between the islands.

We designed a special sample with four areas, each with different island spacing, and used a powerful x-ray holography technique to directly image the magnetic state of every single island as we applied an external field. This allowed us to separate the effects of the islands and the trenches.

Why does this matter?

Our direct imaging experiments, confirmed by our models, revealed that the stray magnetic field from the trench material has a very strong and undesirable influence on how the islands switch. In fact, its effect can be even stronger than the interference from neighboring islands. This shows that creating separate islands isn't enough; the magnetic state of the material between them is also critically important. This provides important information for designing future ultra-high-density storage systems.

Demonstration of the Thermal Stability Advantage of Advanced Exchange Coupled Composite Media

Sylvia H. Florez, Yoshihiro Ikeda, Tyler Hennen, Frank Q. Zhu, Kentaro Takano, Bruce D. Terris

Journal of Applied Physics

May 2014

We tested the magnetic layers in advanced hard drive media to find the ideal balance for the coupling strength between them. Our work shows how to best balance making the magnetic bits easy to write while keeping them stable for long-term data storage.

Finding the Magnetic Balance

To keep packing more and more data onto hard drives, the tiny magnetic bits used for storage need to be incredibly stable, but still easy to flip when writing new data. Advanced hard drive media uses complex stacks of different magnetic layers that are physically coupled together to achieve this balance. We wanted to find out exactly how the strength of this coupling affects performance.

We created a series of media samples where we systematically tuned the coupling strength between the layers, from very strong to very weak. We then ran a series of magnetic tests to measure both the writeability and the thermal stability of the bits for each configuration.

Why does this matter?

Our experiments showed that while weakening the coupling makes the magnetic bits easier to write, it can also make them a bit less stable against heat. However, we found that the overall trade-off between stability and the energy needed to write a bit was excellent across the board and even peaked at an intermediate coupling strength. This provides a clear recipe for engineers to design future hard drive media that can be packed to incredible densities while remaining reliable for decades.

Magnetic Cluster Size “Knee” Analysis for Small Grain Continuous Media

Yoshihiro Ikeda, Sylvia H. Florez, Frank Q. Zhu, Kentaro Takano, Hoa Do, Tyler Hennen, Bruce D. Terris

IEEE Transactions on Magnetics

November 2012

We found that when you shrink the magnetic grains in hard drive media below a certain point, they start to clump together magnetically, which hurts performance. Our work shows how to manage this issue, which is important for developing future high-density drives.

The Problem with Small Grains

The main way to increase hard drive capacity is to shrink the magnetic grains that store each bit of data. But there's a catch. As the grains get smaller, the non-magnetic boundaries between them also get thinner, which can cause them to become too strongly coupled. When this happens, they stop acting like individual bits and instead behave like large 'magnetic clusters,' which increases the media noise and makes it harder to read and write data reliably.

We analyzed various types of magnetic media with different grain sizes and levels of internal magnetic coupling. We measured the 'magnetic cluster size' to see how these groups of grains behave as the physical grain size shrinks.

Why does this matter?

Our analysis identified a critical 'knee' in this behavior: for highly coupled media, the magnetic cluster size suddenly shoots up when the physical grains get smaller than about 8 nanometers, hurting the drive's signal-to-noise ratio. By using media with lower internal coupling, we can push this problematic knee down to even smaller grain sizes. This is key for designing the next generation of hard drive media that can support terabit-level storage densities.