Neuromorphic photonic systems with lasers
Lead Research Organisation:
University of Strathclyde
Department Name: Physics
Abstract
This project seeks to develop novel neuromorphic photonic systems inspired by the powerful computational capabilities of the brain for information processing applications at ultrafast speeds. The project will develop photonic neuronal models and systems using advanced semiconductor lasers as main components. Remarkably, semiconductor lasers can produce dynamical responses analogous to those of biological neurons but several (up to 9) orders of magnitude faster. Moreover, semiconductor lasers are ideal for novel photonic neuronal models as they are compact, offer fast operation speeds and allow integration into photonic circuits and networks. This project will therefore contribute to advance our research vision in brain-inspired photonics for novel paradigms in ultrafast neuromorphic computing going beyond classical digital systems.
The objectives of the project will include the development of novel photonic neurons using different advanced semiconductor laser structures (e.g. Vertical Cavity Surface Emitting Lasers, Nanostructure Lasers). Subsequently, this programme will investigate the scaling of individual photonic neurons into networks with multiple interconnected nodes. This project will also deliver prototypes of photonic neuronal networks/circuits with brain-inspired connectivity. This project aims also at delivering prototypes of neuromorphic photonic systems able to perform information processing tasks at ultrafast speeds using both individual and interconnected photonic neurons. This project will therefore advance the emerging field of neuromorphic photonics offering exciting prospects for functional neuro-inspired ultrafast systems for applications across diverse fields (e.g. computing, optical communication networks, data analysis, etc).
The objectives of the project will include the development of novel photonic neurons using different advanced semiconductor laser structures (e.g. Vertical Cavity Surface Emitting Lasers, Nanostructure Lasers). Subsequently, this programme will investigate the scaling of individual photonic neurons into networks with multiple interconnected nodes. This project will also deliver prototypes of photonic neuronal networks/circuits with brain-inspired connectivity. This project aims also at delivering prototypes of neuromorphic photonic systems able to perform information processing tasks at ultrafast speeds using both individual and interconnected photonic neurons. This project will therefore advance the emerging field of neuromorphic photonics offering exciting prospects for functional neuro-inspired ultrafast systems for applications across diverse fields (e.g. computing, optical communication networks, data analysis, etc).
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/N509760/1 | 01/10/2016 | 30/09/2021 | |||
| 1960309 | Studentship | EP/N509760/1 | 01/10/2017 | 31/03/2021 | Joshua Robertson |
| EP/R513349/1 | 01/10/2018 | 30/09/2023 | |||
| 1960309 | Studentship | EP/R513349/1 | 01/10/2017 | 31/03/2021 | Joshua Robertson |
| Description | One of the key findings associated with this award was the investigation and discovery of semiconductor laser devices as suitable artificial neurons. In this finding we investigated and developed artifical neurons using Vertical-cavity surface emitting lasers (VCSELs). We have demonstrated that these devices have the capability to exhibit neuronal spiking dynamics and that we can control the activation and inhibiiton of these dynamics in different ways. The controllable activation and inhibition of fast neuronal spiking dynamics has been achieved using optically-injected intensity-modulated light from another optical source as well as using bias modulations in optically-injected devices. These techniques developed though this award allow us to control the number of neuronal spiking dynamics produced in our devices and hence help us generate many desired spiking signals. The investigation into these devices has also shed light on other behaviours (other than spike firing) that are shared by both VCSELs and biological neurons. Behaviours such as input thresholding and refractory period have been realised in VCSELs through this award which yet further demonstrate the suitability of these laser devices as candidates for artificial optical neurons. These behaviours, that are characteristic of biological neurons, have helped to expand the functionality we can demonstrate using our artificial VCSEL neurons. A second key finding discovered through this award was the level of interconnectivy achievable in circuits of VCSEL devices. We have demonstrated that multiple VCSEL devices can be connected in cascadable arrangments where one spiking VCSEL can trigger neuronal spiking dynamics in the next. Cascadle networks of up to three devices have been achieved and results on the communication of spiking and inhibiting signals in diverging networks (1 VCSEL into 2 VCSELs) have been completed. Similarly the replication of brain-like neuronal connections, such as autaptic (self-feedback) connections and mutually coupled neurons, have been tested. Through these demonstrations we were able to show early demonstrations of the storage of neuronal spiking dynamics in memory as well as the ability to trigger spiking dynamics controllably within the memory cycle. The interconnectivity of these devices is important as neural networking plays a large role in brain-inspired spike-based computing. Through this award work will continue on the expanding of these interconnected networks of artificial neurons. Further key findings of the work supported by this award would be the spike-based functionality achieved using these VCSEL devices. The investigation of these devices has given rise to all optical analog to spiking format conversion, where digital binary signals fed to the VCSEL are converted into optical spikes. In our demonstrations conversion speeds have reached 1Gigabit/s using return-to-zero (RZ) and nonreturn-to-zero (NRZ) coding formats, providing a new approach utilizing simply 1 VCSEL neuron. Another functionality we have demonstrated is that of the neural circuitry in the retina. Using an arrangment of 3 cascaded VCSELs we demonstrated the emulation of 3 layers of neurons in the retina, the photoreceptors, bipolar cells and retinal ganglion cells. We demonstrated that our system of VCSELs could convert incoming graded potential signals into spiking signals in a demonstration of both analog to spike conversion and device interconnectivity. Also through this award we have discovered a VCSELs ability to integrate (sum together) multiple time delayed inputs. This behaviour, that is key to a neurons ability to recieve multiple inputs in a network, has lead to a number of functionalities in our devices. Using the integration of inputs we are able to demonstrate small circuits capable of AND/OR logic operations where spike firing depends on the number of inputs present. Similarly using a combination of the integration and some weighting we developed a pattern recognition system capable of recognising 4 bit patterns. We have also used integration to perform the convolution of images helping to reveal information about edges and gradients in images. Input integration in our artificial VCSEL neurons has therefore allowed us to realise more application based funtionalities that are helping us move towards task-focused spike-based VCSEL neural networks. |
| Exploitation Route | We hope that the controlled spiking dynamics, the inteconnectivy and the functionality of these artificial VCSEL neurons can be taken forward to help develop fast spike-based systems capabe of different task-based applications such as pattern/image recognition, decision making and information processing. The outcomes of this work should be used to help discover if VCSEL neurons can be used to perform new tasks currently not performed using optical spike-based systems. The outcomes of this work could be used to create individual scalable aritifical VCSEL neuron nodes that can be combined to any network configuration. If established as truely controllable and scalable network nodes we believe that these devices could be used by any reasearcher interested in performing information processing tasks using fast optical spikes. |
| Sectors | Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software) |
| Description | Neuromorphic Dynamics in Optically Injected Lasers |
| Organisation | University College Cork |
| Department | Tyndall National Institute |
| Country | Ireland |
| Sector | Academic/University |
| PI Contribution | In this collaboration PhD students from both groups attended each other institutes for a period of a week. During our research visit to our collaborator, experimental work was conducted in the laboratory on the optical injection of edge emitting quantum dot laser sources. These devices produced excitable dynamics upon optical injection and during our experimental work we helped to controllably activate these dynamics. We helped implement a phase modulator into the optical injection of the quantum dot devices and demonstrated that the controlled firing of these excitable dynamics was achievable. We produced activation efficiency graphs to verify the threshold of these excitable dynamics and analysed results at the time of the research visit in order to obtain a dataset that covered the activation of both square pulses and square dropouts. |
| Collaborator Contribution | During our collaborators' visit to our institute, experimental work on the interconnectivity of spiking VCSEL devices was conducted. The collaborating partner helped design, construct and run an experiment where a single spiking VCSEL communicated dynamics to another 2 VCSEL devices in a 1-into-2 configuration. The collaborator helped select and allign VCSEL devices as well as conduct the experiments. The work completed with the collaborator helped to demonstrate, as a proof of concept, that the VCSELs could achieve diverging network interconnectivity. |
| Impact | The collaboration has resulted in a single journal publication. The publication titled 'Neuromorphic dynamics with optically injected quantum dot lasers', can be found in The European Physical Journal B, volume 92, Article number: 197 and was published in 2019. The work presented in the article relates to the work we completed on our visit to our collaborators. |
| Start Year | 2018 |