Memory Cells for Future Computers
They are the top candidates when it comes to making computers and smartphones more powerful and above all more energy efficient: resistive memory cells. New findings by Jülich researchers could help to establish these nanoelectronic components as data storage units over the next few years. In the more distant future, they may even serve as artificial synapses modelled on biological neurons.
Jülich's first supercomputer constructed in the 1980s had a storage capacity of 16 megabytes. Today’s USB memory sticks have many times this capacity and they also fit into your pocket.
We have become accustomed to the fact that more and more information can be stored on increasingly smaller spaces. However, it cannot be taken for granted that this will go on forever: the miniaturization of existing data storage technologies may be approaching its natural limits. Moreover, none of the available data storage technologies is an all-rounder: ideally, they should work at breakneck speed but be energy-efficient at the same time, they should archive data permanently, and, in addition, they should be inexpensive. Information stored on the DRAM module of a computer’s working memory, for example, will disappear when the power is turned off. Hard drives and flash devices don’t have this problem, but they are comparatively slow.
ReRAM: New kind of data storage
For these reasons, researchers in science and industry worldwide are working on a new type of data storage referred to as ReRAM (resistive random access memory).
"In principle, ReRAMs should be able to store data on an even smaller space than flash memories, for example, and they should also be able to do so with far less power," says Prof. Rainer Waser, director at the Peter Grünberg Institute, Forschungszentrum Jülich, and researcher and professor at RWTH Aachen University. He and his colleagues cooperate closely with companies such as Intel and Samsung Electronics and are considered pioneers in the area of resistive switching elements.
Resistive memory cells store the two basic elements of all computer languages – "zero" and "one" – in a way that is fundamentally different to hard drives or flash devices. In a hard drive, information is stored on the magnetic layer of a rotating disk, while flash devices store bits in the form of electric charges on a special transistor. A resistive memory cell, in contrast, saves a bit by using its electrical resistance, which can be switched between high and low values, – and the memory cell retains its state even when the external voltage is switched off.
Using ions instead of electrons
The resistive memory cells currently being produced and studied in laboratories all over the world have an edge length shorter than 30 nanometres, i.e. 0.00003 mm, or even smaller.
A common configuration of these cells consists of three thin layers of materials with different functions. The material in the middle, the electrolyte, is like the filling between two slices of bread in a sandwich. One of these "slices" is an active electrode made of metal, for example copper, and the other is a ounter electrode made from a chemically inert material such as platinum (see graphic). When a voltage is applied, positively charged copper ions are released from the active electrode and migrate to the counter electrode, where they gain electrons and form elementary copper atoms once again. The atoms form a narrow pathway through the electrolyte – experts refer to this as a filament. Once the filament has formed and an electrically conductive contact has been established between the two electrodes, the resistance of the entire cell is low and it is in the ON state. This corresponds to the "one" in computer language. The cell remains in this state until an appropriately high voltage of reverse polarity is applied. This causes the filament to dissolve and the cell resistance increases to a high value. The cell is now in the OFF or "zero" state.
Unlike conventional technologies, this way of storing information is based on ions. Ions are primarily responsible for forming and dissolving the filament, the switching processes, and therefore for information storage. This gives the technology a significant advantage over flash memory devices that store electrons, since the latter can easily find a way to break free, causing the information to be lost. Ions are easier to manage in comparison to electrons and they are therefore more suitable for reliable information storage.
In the past, scientists described resistive memory cells using "memristor" theory (memristor is a portmanteau word from memory and resistor). According to this theory, resistive memory cells are passive components just like capacitors, coils, and resistors, all of which neither amplify a signal nor have a control function. An important characteristic of passive components is that no current flows through them unless a voltage is applied. Conversely, no voltage can be measured without current flowing.
Research from Jülich, however, theoretically deduced that this is intrinsically different in resistive cells. The cells produce a voltage like a tiny battery. The scientists want to use this discovery of a battery voltage in ReRAMs to optimize the reading out of information from these cells.
Experts previously assumed that electric current is needed to determine whether resistive cells are in the ON or OFF state. However, this current could potentially alter the state of sensitive cells. The battery voltage, in contrast, can be measured without current in a nondestructive process. The researchers have already filed patent applications for the relevant methods. “Furthermore, the existence of a battery voltage must also be taken into consideration when connecting resistive memory cells or developing reliable ReRAMs,” says Dr. Ilia Valov, an electrochemist in Waser’s research group.
Integration of ReRAMs into existing semiconductor technology
Other scientists working with Rainer Waser in the Jülich Aachen Research Alliance are already building larger units from resistive memory cells and using computers to simulate the integration of ReRAMs into existing semiconductor technology. Panasonic has begun integrating ReRAMs into their microcontrollers. Despite this, Valov’s colleague Prof. Regina Dittmann is also convinced that further basic research into resistive memory cells is necessary to improve characteristics such as the durability and switching speed of ReRAMs. "Industry is continuously confronted with challenges in this respect that can only be overcome by improving our understanding of the basic principles,2 says Dittmann.
The scientists now have access to a new laboratory at Jülich’s Peter Grünberg Institute – the Oxide Cluster – for their fundamental experiments. The facility allows scientists to produce layers of materials and resistive memory cells and to use the latest methods in microscopy and spectroscopy, for example to observe the atoms and electrons during switching processes and therefore "at work" without the materials leaving the ultrahigh vacuum. This is important because contact with air would influence the processes that occur at the surface of these materials.
Parallels to synapses in the human brain
In principle, resistive memory cells do not just switch between two resistance values but rather between several. In other words, they can also assume intermediate states between "zero" and "one". This provides a good basis for creating computer systems capable of learning according to the model of synapses the interfaces between cells in the biological nervous system.
"There are also a number of parallels between how biological synapses and resistive components function," says Dr. Susanne Hoffmann-Eifert from Waser’s research group. For example, the functioning of both synapses and resistive cells is based on the movement of ions. Another similarity is that a connection between human neurons becomes stronger and more efficient the more frequently and intensively it is used.
"The more current that is transmitted through the conductive filaments in resistive memory cells, the stronger they become," says Hoffmann-Eifert. One day, this effect could help us to create computers capable of deviating from their programs independently if a connection is unexpectedly intensively used.