Periodic Reporting for period 1 - TRANSIONICS (Solid-State Ionics Synaptic Transistors for Neuromorphic Computing)
Reporting period: 2022-06-01 to 2023-11-30
In the recent years, synaptic transistors have gained attention for analog computing solutions. They aim at overcoming the Von Neumann bottleneck, a limitation in traditional digital computing caused by the separation of data memory and processing units. Synaptic transistors mimic the behavior of biological neurons improving the energy efficiency and processor performance for novel brain-inspired computing, such as neuromorphic computing and artificial intelligence. Among the possible solutions proposed in literature, Electrolyte-Gated Transistors (EGTs) have emerged as a very promising approach. EGTs are essentially conventional Field-Effect Transistors (FETs) where the gate dielectric is replaced by an ion-conducting electrolyte. However, current EGTs rely on unstable electrolytes challenging to integrate, such as liquids or polymers. Alternatively, more robust oxide-ion conductors operate the EGTs at elevated temperatures (above 200 ˚C) not compatible with standard microelectronics. To address this, TRANSIONICS has proposed a three-terminal thin-film EGT based on a ceramic electrolyte able to work at low temperatures developed by the team during Prof. Tarancón’s ERC Consolidator Grant, ULTRASOFC, and a mixed ionic-electronic conducting channel. In this device, a gate bias is applied to change the oxygen stoichiometry of the channel by pumping oxygen ions across the electroyte. In such a way, the conductance of the channel is controlled by in an anologic fashion, giving rise to a non-volatile multistate transistor. This technology is compatible with microfabrication processes, allowing for low-cost and large-scale production.
The main technical objective of the TRANSIONICS project is to create a minimum value product on silicon substrate, and to develop oxide-ion electrolytes based on functional substrate scalability assessment. In addition, the definition of the market uptake strategy is the main business-oriented goal of the project. In that regard, the management of the intellectual property, first contacts with potential users and development of a business plan are key activities.
The main technical objective of the TRANSIONICS project is to create a minimum value product on silicon substrate, and to develop oxide-ion electrolytes based on functional substrate scalability assessment. In addition, the definition of the market uptake strategy is the main business-oriented goal of the project. In that regard, the management of the intellectual property, first contacts with potential users and development of a business plan are key activities.
During the project, several generation of devices have been fabricated. The first one have been fabricated on ceramic substrates and had large dimensions (thousands of microns). This first generation of EGTs have been tested at 300 ˚C since the large dimensions give rise to significant electrolyte resistance. However, the positive results have been relevant to demonstrate the possibility of varying almost 3 order of magnitude the channel’s conductivity, a very linear and symmetric synaptic behavior induced by current pulses and a non-volatile character of the states. With the aim to reduce the operating temperature below 100 ˚C, a microfabrication route employing large-scale fabrication techniques has been demonstrated, allowing a new generation of micro and nano size devices. Initially, oxide-based single crystal substrates have been used for the EGT prototypes. However, first attempts to develop the device on silicon substrate have been successfully achieved, since this was the ultimate goal to enable compatibility with existing microelectronics manufacturing processes.
Regarding the business development, a market assessment has been also conducted, focusing on the neuromorphic computing sector. This, along with insights gained from several conversations with potential end-users, led to the identification of the market entry strategy and business model, both of which have been collected in a preliminary business plan.
Regarding the business development, a market assessment has been also conducted, focusing on the neuromorphic computing sector. This, along with insights gained from several conversations with potential end-users, led to the identification of the market entry strategy and business model, both of which have been collected in a preliminary business plan.
The technical outcomes of TRANSIONICS are currently being compiled into a scientific paper that will be published shortly. Moreover, the intellectual property of the project has been protected through a patent, which is in the process of validation in several countries of interest. TRANSIONICS and other breakthrough synaptic transistors technologies are expected to achieve a large interest from the Artificial Intelligence (AI) community. Currently, AI relies on brute force to exploit its potential. However, this method is not scalable and challenges such as heat, data or cost constrains are anticipated, all of which are associated to the limited capacity of the semiconductor industry to overcome Moore’s law and the incremental costs related with the performance improvements.
Neuromorphic computing solutions, such as TRANSIONICS technology, unlike traditional von Neumann architecture, allows for massive and low energy parallel processing, similar to how neurons in the brain process information simultaneously. This enables faster and more efficient computation for AI algorithms. These technologies could represent 20% of all AI computing & sensing by 2035.
Althouth TRANSIONICS has demonstrated the great potential of the single synaptic transistors to be applyed for neuromorphic computing, it will still be necessary to evolve the technology from the sinlge device into a neural network of transistors, in order to show its functionality and scalability for deep learning accelerators in real applications. This tehcnological advancements and scalability within the semiconductor industry will be pursued in future research projects.
Neuromorphic computing solutions, such as TRANSIONICS technology, unlike traditional von Neumann architecture, allows for massive and low energy parallel processing, similar to how neurons in the brain process information simultaneously. This enables faster and more efficient computation for AI algorithms. These technologies could represent 20% of all AI computing & sensing by 2035.
Althouth TRANSIONICS has demonstrated the great potential of the single synaptic transistors to be applyed for neuromorphic computing, it will still be necessary to evolve the technology from the sinlge device into a neural network of transistors, in order to show its functionality and scalability for deep learning accelerators in real applications. This tehcnological advancements and scalability within the semiconductor industry will be pursued in future research projects.