In this research area, the aim is the simulation and fabrication of phase-change material on-chip photonic memories. We will propose novel architectures, with the view to develop non-volatile, sub-ns and pJ scale memory devices.
If you are interested about this topic, contact Mr. Emanuele Gemo.
Phase-change materials (pcm) have been discovered almost 60 years ago. The seminal work of S. Ovshinsky led to an ever increasing research over this class of materials, due to the peculiar stability of different solid phases at room temperature, as well as the diverse electric and optical phase-dependent properties. Admirably, Ovshinsky has predicted many different applications, some of which are being and proposed in these days, each one pushing renovated interest in this field.
Out of the many technological applications, rewritable DVDs are the most common ones. They do exploit a thin layer of pcm which can be alternatively switched from one phase to the other where the writing laser spot delivers the appropriate power pulse, leaving a readable mark corresponding to the information bit. The reflectivity contrast between the written and non-written pcm is the key feature. More recently, Intel and Micron Technology proposed a solid-state device which once again exploits pcms, the 3D X-Point Optane memory, which (probably!) exploits the different conductivity of pcms phases.
And research is going even beyond the estabilished von Neumann computing architecture, where processing unit and memory are distinct and separated units. For example, IBM proposed one of the first non-von Neumann computing device, where computation is performed right in memory. And once again, they rely on phase-change materials. Outstanding, but..
Can computing be faster?
One approach might be the one where light replace current. Light travel fast, way faster then charge carriers. If losses are present, they are not always linked to heat generation, which is a major drawback for nanometric components in silicon technology. And, most of all, a single waveguide can propagate multiple wavelengths at the same time, thus providing a built-in broadband channel between components. One proposed configuration, can be found in this article. It exploits light to enhance the communication speed between cpu and memory, thus downsizing the so-called von Neumann bottleneck issue. Still, it relies on electronic to optic converters: and from this point, we can talk about the role of integrated photonic memories.
A on-chip integrated photonic memory would totally replace the need of optic to electronic converters, so making a step towards the complete replacement of voltage signal with light signal in computing devices.
A phase-change integrated photonic memory design is rather simple to visualise: a thin layer of pcm placed (called optical cell) on top of a waveguide. Due to the evanescent field, the travelling light pulse interacts with the pcm thin layer, being partially absorbed and locally increasing the optical cell temperature.
The different extinction coefficient of the involved material phases provides a change-in-readout (modulation of the transmission) which, like the DVD reflectivity modulation, tells whether we have a state 0 or a state 1. And that’s how we read a memory level.
Writing is performed via through the same optical path, providing a powerful pulse. A single short powerful pulse, due to the absorption, push the optical cell temperature above melting temperature; after the pulse, the pcm doesn’t have enough time to recrystallise, and thus, quickly freezes in the glassy phase. This procedure is called melt-quench process. When we need to retrieve the former state, a longer, less powerful pulse, is provided, which increases the optical cell temperature above a threshold called crystallisation temperature.
One of the main objective of our work is the study of the principles behind the photonic memory functioning.
We adopt finite element methods to calculate the propagation of the e.m. pulse through the waveguide and the resulting temperature distribution.
We are also building the most convenient method to achieve a finely resolved phase-transition model to be used in combination with the former, to track the evolution of the pcm phase as a result of the provided pulse.
Our aim is to develop a reliable tool capable to deal with all the involved elements and to produce a realistic behavioral model for each photonic memory architecture. At the same time, we’re exploring different architectures, with the goal to increase light-matter interaction and therefore allowing to increase the memory performances, in terms of power efficiency, operative maximum frequency, and modulation of signal (which is also related to how many levels a single cell can reliably hold).