Project progress

First report 2019:

A physical mask for photolithography was designed comprised of two sections, the individualized memristor devices and the crossbar array, both a positive and negative mask were made to allow for flexibility in the choice of patterning methods and deposition materials. Fabrication of memristor devices using PVD and ALD techniques was done. Material selection was based on the same semiconductor and dielectric that are commonly used for amorphous oxide thin film transistors (TFTs) technology. Thus, various compositions of IGZO, ZTO and a bilayer structure of Cu2O-Al2O3 were studied as switching network. Different electrodes such as Pt (Inert, cost intensive) and Mo (reactive contact, adaptable with TFTs technology) were tested in different device configurations (patterned by photolithography and shadow masks). Both potential applications of neural networks and RAMs are investigated through material optimization meaning that retention time and memory window are of vital importance. Optimization of devices comprised control of oxygen content and thickness. Design of a bilayer memristor structure using ALD processing (Al2O3 on top of Cu2O) showed a multi-level resistive switching characteristic which can be applied for RAM applications. The memristor devices based on IGZO revealed a reproducible analog pinched hysteresis in the current-voltage loops with high yield for all electrode areas even as small as 4 µm(no device-to-device variation). However, the memristors do not show a good retention time at high temperature. Therefore, they are more suitable for neuromorphic systems. The IGZO TFTs fabricated via combustion method were investigated to be presented for fully solution-processed memristor-transistor integration. Here the dielectric is Al2O3, an optimal choice to simplify the processing steps combined with a solution processed filamentary RRAM device based on an oxygen deficient Al2O3 switching layer.

A behavioral modelling approach, based on data driven methods, was engaged. Starting from a Voltage Threshold Adaptive Memristor (VTEAM) model of a memristor, as a development basis, different types of recursive neural networks (RNN) were designed in python to approximate the function. A complex signal is generated to cover the space of the VTEAM response model, and the pairs of excitation/response were used as training data for the RNN. Further, the resulting RNN was implemented in VerilogA for circuit simulation. Tests were made using Spectre/Cadence spice simulator and the results were successful. Further, a novel type of neural network for differential equation solving was studied and used as a model basis for memristors. Preliminary results show that this approach is very effective (better than RNN). To demonstrate the effectiveness of both methodologies, the characterization of real fabricated memristor devices has started. An electronic circuit was conceived to allow the excitation of the real memristor with complex pulsed-shaped signals and the measurement of the response. These will be used as new training data. 

 Second report 2020:

One of the interesting features of the amorphous oxide memristive device is the presence of both filamentary and area dependent resistive switching characteristics based on voltage bias polarity. This property is shown in the recent published paper on zinc-tin oxide (ZTO) memristors. The area-dependency or 2D resistive switching can be further realized in reliable neural network systems and commercialization of the memristor-based circuits. We note that this important feature is not highly investigated in the memristor research community. For the area-dependent resistive switching properties, amorphous indium-gallium-zinc-oxide (a-IGZO) with molybdenum contacts as both top and bottom electrodes with the area down to 4 μm2 were also fabricated. Due to the existence of a thin intermixed molybdenum oxide layer (4-5 nm) at the interface of the bottom contact, Schottky diode-like characteristics at the pristine state can be obtained. We believe that the oxygen gradient then is easily produced without a need of expensive contact materials such as Pt. These devices also show area-dependent and analog resistive switching properties. Several synaptic functions such as synaptic potentiation and depression as response to programmed pulses, short to long term plasticity transition and “learning experience” was also tested on the current devices.

TFT/memristor PDK support for the latest IC-CAD version was added. The DRC and LVS was updated to the current manufacturing process, including parasitic extraction, as well as joint memristor/TFT device design possibility (Pcell + layout rules). Memristor modeling is an ongoing process, several measurements were made, and different strategies applied, namely RNN fitting, which was not able to capture the full complexity of the device, however it gave a very fitting on DC sweeping data. Further ODE Network (Ordinary differential equation network) was employed which has shown better performance in keeping track of the device's state. The attempt now is to move forward to a hybrid model (Physical + ODE Network) that so far is considered the best alternative.

 Third report 2021:

Scientific outputs and results (Task 2, 3, 4, 5):

In the third year, four articles were published in international journals, several conference communications and three master theses in line with NeurOxide project were concluded.

Details to the work done is below:

Under the goal of activity 2 (memristor fabrication: solution processing), an article related to applying printing technique for the first time to fabricate memristors is published 1. In addition, a master thesis related to development of solution processed IGZO memristors is fulfilled (Raquel Azevedo Martins, thesis will be available on NeurOxide website).

Under the goal of activity 3 (Modeling and design platform), an article related to behavioral model of memristor is published2. A neural network approach towards a generalized modeling of device behavior is demonstrated. It presents an artificial neural network learning approach to resistive switching modelling of amorphous IGZO memristors. A normalized root-mean-squared error (NRMSE) of 5.66 × 10−3 is achieved with a [2, 50,50 ,1] network structure, representing a good balance between model complexity and accuracy.

Under the goals of activity 4 (Thin film transistor and memristor integration), one-transistor-one-memristor (1T1M) crossbar array was built on glass substrate. Both thin-film transistor (TFT) and memristor are fabricated at the same level, using the same processing steps and therefore, sharing the same material layers. This proposed layout implies a decrease in total mask count, improved interconnectivity, and drastic cost reduction. Moreover, TFTs as support electronics can be fabricated also at the same level as the crossbars on the neuromorphic chip. This work is ongoing. Active cross bars consist of 1T1M cells and ternary inverters are ready to be tested. A picture of the chip is shown in the attached file.

Under the goals of activity 5 (Logics, ANN prototype), an article is in the review process to be published in journal of APL material3. Here, we present capability of IGZO memristor for ANN system by gaining pattern recognition accuracy of 91.8%, using MNIST handwritten digits dataset (CrossSim platform). Furthermore, design of an imply logic chip with controlling TFTs have been accomplished for in-memory demonstration. This design is currently in the phase of fabrication and is a result of an MSc thesis from Luís Outeiro.

Other achievements:

1. Further explore of material optimization mainly related to build a passive cross bar (with no transistor/cell) is published4. It shows the capability of barrier engineering of ZnO- based devices to build a self-selective memristor device, as a bonus alternative for a high-density chip.

2. The second activity of the Neuroxide project, opened a door of opportunities towards further development of printed memristors following a granted project in 2021 entitled: Supreme-IT [EXPL/CTM-REF/0978/2021]- PI: Jonas Deuremeier, (Members: E. Carlos, A. Kiazadeh, P. Freitas).

3. An article is published5 to validate the potential use of crossbar memristive devices based on amorphous oxide semiconductor (here Zinc tin oxide devices) for a sustainable and energy-efficient brain-inspired deep neural network computation. Furthermore, C. Silva, who developed a master thesis in 2021 in line with the goals of the project (The thesis is online in the NeurOxide website) is currently granted for Phd with the reference 2021.07840.BD. He will be involved to study/explore/design sustainable memristors for artificial neural network applications under supervision of PI of the NeurOxide and co-supervision: J. Deuremeier, W. Zhang (Liverpool John Moores University).

 Poster communications:

1. Presented- Emanuel Carlos, Asal Kiazadeh*, Raquel Martins Rita Branquinho, Jonas Deuermeier, Rodrigo Martins, Elvira Fortunato, “Recent Challenges and the Future of Metal Oxide Resistive Switching Devices” Materials Challenges for Memory - APL Materials 2021, April.  URL:

2. Presented- Maria E. Pereira*, Jonas Deuermeier, Carlos Silva, Pydi Ganga Bahubalindruni, Pedro Barquinha, Rodrigo Martins, Elvira Fortunato and Asal Kiazadeh*, “AOS-based memristive devices towards TFT integration: Materials and challenges”, Materials Challenges for Memory - APL Materials 2021, April. URL:

3. Presented- Carlos Silva, Jorge Martins, Jonas Deuermeier, Maria E. Pereira, Pedro Freitas, Wei Zhang, Rodrigo Martins, Elvira Fortunato, and Asal Kiazadeh*, “Zinc tin oxide memristors for neuromorphic applications”, Euro Nano Forum 2021, May. URL:

4. Presented- M. Pereira, P. G. Bahubalindruni, P. Barquinha, Asal Kiazadeh*, “Memristive devices for neuromorphic applications based on amorphous oxide semiconductor nanoscale films”, 12th Advanced Doctoral Conference on Computing, Electrical and Industrial Systems 2021, July. URL:

5. Presented- Jonas Deuermeier, Maria Pereira, Jorge Martins, Carlos Silva, Philipp Wendel, Dominik Dietz, Andreas Klein, Rodrigo Martins, Elvira Fortunato, Asal Kiazadeh, “An overview of charge-trapping type switching in mixed transition metal oxide Schottky diodes”, E-MRS Spring Meeting 2021 - Symposium G: Solid state Ionics: advanced concepts and devices, 31 May - 4 June. URL:

6. Accepted- M. E. Pereira*, J. Deuermeier, P. Freitas, P. G. Bahubalindruni, P. Barquinha, W. Zhang, R. Martins, E. Fortunato and A. Kiazadeh*, “Tailoring resistive switching characteristics of IGZO-based memristive devices for artificial deep learning neural networks”, MEMRISYS 2021, 4th International Conference on memristive materials, devices & systems, November. URL:


Oral presentations:

1. Presented- V. Tavares, A. Kiazadeh “Memristors for Neuromorphic Computing”, INESC TEC open day, May 4th, 2021.

2. Presented- V. Tavares, A. Kiazadeh “Memristors for Neuromorphic Computing”, DCE, Porto June 28th, 2021.

3. Presented- M. E. Pereira*, J. Deuermeier, P. Freitas, P. G. Bahubalindruni, P. Barquinha, W. Zhang, R. Martins, E. Fortunato, and A. Kiazadeh* “Synaptic characteristics of IGZO-based memristors networks for pattern recognition applications” at 10th International PhD Meeting, Fraunhofer, Dresden, September 2021. URL:

4. Invited talk accepted- Jonas Deuermeier*, Maria Pereira, Emanuel Carlos, Carlos Silva, Rodrigo Martins, Elvira Fortunato, Asal Kiazadeh, “Resistive switching materials and devices” at CIMTEC 2022, symposium CM-2. URL:



1. URL: 

2. URL:

3. Submitted- Pereira, M et al, “Tailoring the synaptic properties of a-IGZO memristors

for artificial deep neural networks”, under revisions in APL materials, October 2021

4. URL:  

5. URL:

 *Corresponding author


Final report 2021:

Scientific outputs and results (Task 4 and 5):
In the fourth year, four articles were published in international journals, several conference communications, book chapter, invited talks, awards and cover images of journals in the scope of NeurOxide project results were achieved.

Under the goals of activity 4 and 5

Successful integration of IGZO-based TFT and memristors in the concept of 1 transistor-1 memristor (proof of concept) was done using PVD processing. This part of work is reported in refence [2], [4]. In addition, a low-cost solution processed IGZO with superior endurance, and retention up to 105 s was reported in master thesis of Raquel Martins and published in 2022 in the journal [3]. The optimized devices can be programmed in a multi-level cell operation mode, with 8 different resistive states. Also, preliminary results reveal synaptic behavior by replicating the plasticity of a synaptic junction through potentiation and depression; this is a significant step towards low-cost processes and large-scale compatibility of neuromorphic computing systems. In this regard, printed IGZO memristors was fabricated, showing interesting results towards reservoir computing application. The work is reported, and it is in the process of reviewing (Submitted article in the appendix file).

PVD-processed 4×4 crossbar array including 1-transistor 1-memristor (1T1M)/cell was also designed, for memristive neural network hardwares (ANN prototype). The a-IGZO crossbar is built on both glass and eventually on a flexible polyimide substrate (biocompatible substrate), the latter enabling IoT and wearable applications. In the novel framework, the thin-film transistor and memristor are fabricated at the same level, with the same processing steps and sharing the same materials for all layers. The 1T1M cells show linear and symmetrical plasticity characteristic with low cycle-to-cycle variability, thanks to the area-dependent analog resistance change. The memristor performs like an analog dot product engine and vector–matrix multiplication in the 4×4 crossbars are demonstrated experimentally, in which the sneak-path current issue is successfully suppressed, resulting in a proof-of-concept for a cost-effective flexible artificial neural networks hardware [5].

A ternary inverter (RRAM prototype) was also fabricated and tested. However, due to depletion operating transistors, the results were not satisfactory and further tuning either in the circuit or processing was required. In this respect, analogue and digital depletion-mode single channel transistor circuits were simulated using n-channel IGZO technology with V t = −0.87 V. A logic family is introduced, suppressing the need for an additional voltage level and level restoring circuitry.  The work is submitted to ISCAS 2023.  The masks are ready for fabrication. Delay in fabrication is currently due to cleanroom closure at CENIMAT for repairing gas pipelines leakages.

Articles and book chapter of the last year of the project:

1. Characterization and modeling of resistive switching phenomena in IGZO devices, AIP Advances12,

2. Tailoring the synaptic properties of a-IGZO memristors for artificial deep neural networks", APL Materials 10 1 (2022): 011113-011113.

3. Emergent solution based IGZO memristor towards neuromorphic applications, J. Mater. Chem. C, 2022, 10, 1991-1998.

4. Flexible Active Crossbar Arrays Using Amorphous Oxide Semiconductor Technology toward Artificial Neural Networks Hardware. Advanced Electronic Materials (2022):

5. Maria Elias Pereira; Emanuel Carlos; Elvira Fortunato; Rodrigo Martins; Pedro Barquinha and Asal Kiazadeh. "Amorphous oxide semiconductor memristors: brain-inspired computation". Book chapter in Advanced Memory Technology. Royal Society of Chemistry. Under revisions.

Conferences and awards:

Awards 2022:

Best Oral Presentation Award: “Flexible 1-transistor-1-memristor crossbar for Artificial Neural Network hardware” at the 11th European School for Young Materials Scientists- 27-28 September, Lisbon, 2022.

Best Contributed Presentation Award: In-memory vector-matrix multiplication using flexible thin-film 1- transistor-1-memristor crossbar" at the 17th International Thin-Film Transistor Conference (ITC 2022), 14-16 September, Surrey, UK, 2022.

Young Researcher award based on oral presentation: "Active crossbar using amorphous oxide semiconductor technology towards artificial neural networks hardware" at the 2022 European Materials Research Society (E-MRS) Conference - Symposium E: Adaptive materials and devices for brain-inspired electronics, 30 May- 3 June, 2022.

Excellent Poster Presentation Award: "Tailoring resistive switching characteristics of IGZO-based memristive devices for artificial deep learning neural networks" at the 4th International Conference on Memristive Materials, Devices & Systems (MEMRISYS, 2021), 1-4 November, Japan, 2021 -Article "Towards Sustainable Crossbar Artificial Synapses with Zinc-Tin Oxide" chosen as Editor's Choice Article in Electronic Materials.

PI of the Project (Asal Kiazadeh) was awarded for 6 years as Assistant researcher at i3N/CENIMAT Recognized by competitive contest of CEEC (only with 5% acceptance in 2021), 4th edition- (ref.2021.03386.CEECIND).

Keynote: Asal Kiazadeh, AI hardware with amorphous oxide semiconductor memristors and transistors, Symposium resistive switching, physics, devices and applications, Nano 2022, Sevilla, Spain.

Invited talk: Asal Kiazadeh, System-on-panel neuromorphic application by using amorphous oxide semiconductor memristors. International conference on frontier materials 2022,

Invited talk: Jonas Deuermeier, Resistive switching materials and devices at CIMTEC 2022, symposium CM-2, 2022.

Oral talk: Guilherme Carvalho, Compact modelling of amorphous oxide semiconductor memristive devices for memristive neural network, IEEE, mini colloquium, 6th Symposium on Schottky Barrier MOS (SB-MOS) devices, Germany, SEP. 7, 2022.

Oral talk: Guilherme Carvalho, Conduction analysis of IGZO memristor for flexible analog memristor-based neural networks. EMRS, Synthesis, processing and characterization of nanoscale multi-functional oxide films VIII and 6th E-MRS & MRS-J bilateral symposium30 May- 3 June, 2022.

Oral talk: Asal Kiazadeh, Memristor-based Artificial Neural Network (ANN) hardware based on flexible semiconducting technology, CIMTEC: FH - 4th International Conference, 2022.

In overall, a significant outcome was achieved in the scope of Neuroxide including 16 articles (2- submitted) related to the project activities and more 5 articles which benefits from project results (citing NeurOxide project), can be acknowledged in web of science: ( 2 book chapters: “Amorphous oxide semiconductor memristors: devices, models, and circuits, in progress for publication 2022 November", Book: Advanced Memory Technology, Royal Society of Chemistry 2022. 2. "Flexible and transparent RRAM devices for System-On-Panel (SOP) application, Book: Advances in non-volatile memory and storage technology", 2nd Edition, Elsevier 2019, 10.1016/b978-0-08-102584-0.00014-0.

Master thesis: 8 master thesis was fulfilled in the scope of neuroxide project, in which 4 of them are available in RUN repository and added with URL in the platform of final report. The ones which currently have no URL are: 1. Ricardo de Gregório Nogueira 2020- 2. Afonso Monteiro 2020- 3. Carlos Silva 2020- 4. Luís Diogo de Almeida Outeiro 2021- 5. Raquel Azevedo Martins 2021 (all already mentioned in previous reports). Furthermore, a second exploratory proposal as a logical follow-up was funded in 2022. (OPERA: Opto-electronic resistive switching devices for artificial intelligence- Asal Kiazadeh as PI, the project will start in 2023, brief relation of OPERA to NeurOxide: AOS material is applied targeting the activation of neurons by optical stimulus under spiking neural network model.)

Another significant outcome was establishment of research team consist of 4 PhD students funded by FCT, who are progressing the project goals further beyond the defined activities:1. Optoelectronic memristive devices based on oxide semiconductors for the next generation of information technology (Granted by FCT, Ref. 2020.08335.BD), DCM, i3N/CENIMAT, Maria Pereira (former scholarship holder of the project). 2. Metal Oxide memristive-based devices for large-scale artificial neural networks (ANNs) applications (Granted by FCT, Ref. 2021.07840.BD), DCM, i3N/CENIMAT and Nanoelectronics Department of Liverpool John Moores University, Carlos Silva.3. Printed metal oxide memristors: a demand for high energy efficiency in neuromorphic applications (Granted by FCT, Ref. 2022.13773.BD), DCM, i3N/CENIMAT, Raquel Martins. 4. Integration of thin film memristors and transistors for a neuromorphic hardware computational framework, (SFRH/BD/144376/2019), Guilherme Carvalho, INESC TEC, DCM and CENIMAT/I3N.