Abstract: This document records the manufacturing processes and strategies for artworks in the H.Om.E Project, based on dye-sensitized solar cell (DSSC) technology. Dye-sensitized solar cells are a relatively low-cost and easy-to-produce third-generation solar technology. They are glass-based and semi-transparent, capable of being naturally or chemically dyed, and can be customized with patterns through screen printing techniques. This paper proposes using DSSC as a carrier for the concept of generative memory and explores the idea of utilizing its photoelectric properties and variational auto-encoder (VAE) for sound art installations. Finally, it discusses the potential of integrating these cells into textiles and introduces the material and conceptual connections between dye-sensitized solar textiles and memristor matrices.

Keywords: dye sensitized solar cell, tio2, memory philosophy, VAEs, RAVE, memristor

1. Introduction

The H.Om.E Project is composed of three previous projects: Tribal Against Machine, Mind of A Greenhouse, and the I_C Project. The nature of these projects has directed the research team towards developing artifacts in the realms of wearable devices, electronic textiles, and solar panels. Therefore, this study focuses on the shared structures between electronic engineering and textiles, such as the similarity between matrix arrays and weaving structures. Another connection between this study and wearables and textiles comes from the Extended Mind Theory (EMT), which posits that external resources situated outside the skull can become part of the cognitive machinery under certain conditions, provided they are portable, easily accessible, and automatically trusted (reliable). This idea resonates with the “Fuzhi” theme proposed in the Tribal Counter Machine, which is about “how to wear memory” — how to combine wearable device technology with traditional clothing to achieve the preservation of traditional attire.

To echo these concepts, this paper presents several prototype concepts, some of which have already entered the implementation stage, while others are still in the conceptual phase. The following records the current development status of these prototypes and provides a brief assessment of their presentation. These prototypes offer the potential for future implementation of physical art installations that connect the concepts of “generative systems and extended mind as abstract transformation models to link two cultures” mentioned in “Generative Systems and Extended Mind as Abstract Transformation Models Connecting Two Cultures” (in publication). Section three of this paper introduces the connection between these conceptual prototypes and the project concepts, as well as the iterative development. These records focus on conceptual thought changes but omit the details of production methods.

2. Project concept — generative memory

In “Generative Systems and Extended Mind as Abstract Transformation Models Connecting Two Cultures” (in publication), the use of generative AI, such as variational autoencoders (VAEs), is discussed within the framework of heritage preservation and textile design. The significance and methods of using these models as abstract transformation models in cognitive neuroscience are explored, and the concept of “generative memory” is proposed as having parallel roles in both cognitive neuroscience and the arts. The article combines the Extended Mind Theory (EMT) and Ingold’s concept of the network of things to describe a planetary relationship in an abstract and perceptual network.

3. Iterations of prototypes

3.1 Core rope memory

Figure 1. Core Memory Module, A 32 x 32 core memory plane storing 1024 bits (or 128 bytes) of data. The small black rings at the intersections of the grid lines, arranged in four squares, are ferrite magnetic cores.

In this study, the first thing that came to mind related to the concept of “heritage preservation” and weaving technology was the famous magnetic core memory in the electronic textile art field. This is the most closely related computing technology to weaving technology in the history of computer development, following the invention of the Jacquard loom. We can refer to Daniela K. Rosner’s design methods, which involve a deep exploration of the heritage of design and an expansion of the definition of “design” to include long-overlooked practices. This reinterpretation of design history aims to enhance current design methods by integrating these rediscovered practices. Her research is particularly notable for its feminist technoscience perspective, where she examines the intersection of craft and computing. For example, she explores how traditional craft practices, such as those used by the “little old ladies” who wove wires for NASA’s Apollo missions, contribute to modern hardware manufacturing. This fusion of history, theory, and personal narratives demonstrates how craft and hardware manufacturing can influence and transform each other, offering new insights and methodologies for design. Her book, “Critical Fabulations,” proposes redefining design as not only investigative and activist but also personal and culturally situated. This redefinition challenges the dominant paradigms in the field of design and aims to rework design practice from within by acknowledging and integrating the long-silenced stories of craftsmanship and technical work. Rosner provides an excellent model for the manufacturing methods of artifacts in the H.Om.E Project. However, due to the high labor costs required to manufacture magnetic core memory, we abandoned the idea of recreating this technology in practice and instead pursued more contemporary computing technologies.

3.2 Fog collector in the Atacama Desert

Another previous project of H.Om.E, I_C Project, curated by Maria Jose Rios. This project aims to rediscover ancient culture from the perspectives of textile design and data visualization, including aspects such as astronomy, writing, and symbols. The project originated in the Andes Mountains of Chile, which is also the location of the ALMA Observatory. Supported by ALMA’s technology and art grant, Shih Wei-Chieh began participating in the project in 2022 and responded to this spirit in Taiwan with a satellite project, I_C X ALMA X Laser Dye. This initiative aims to recreate the dark constellation culture of the Mapuche civilization using contemporary astronomical data.

Fog collector technology is a process of collecting water from the air through the condensation of water vapor, which can be an effective method of obtaining water in areas with limited freshwater resources. The Atacama Desert is one such area, known as one of the driest places witht the clearest sky on Earth which is why ALMA was built here. There are many atmospheric water collection technologies in the Atacama Desert, one of which is using fog collectors. These are large mesh or net structures that capture water droplets from fog. Fog collectors are usually set up on hills or other areas where fog is likely to occur. They then collect and store the captured water, making it an effective way to obtain water in arid regions.

Figure 3. (A) Study site locations in the coastal Atacama Desert indicated with red dots (standard fog collector). Blue dots indicate the closest urban areas to the study sites. Standard fog collectors along with weather stations in Alto Patache (B) and Falda Verde ©.

Another study demonstrated an interesting correlation between weaving structures, geographical locations, and ecosystems (Figure 3). Researchers conducted experiments at four different fog collection sites along the central coast of California. These sites have diverse ecosystems and geographical attributes. The results indicated that matching different mesh shapes to specific geographical environments is crucial for enhancing the effectiveness of fog water collection (Fernandez 2017).

Another study assessed the fog collection capability at two locations in the Atacama Desert (Alto Patache and Falda Verde). Alto Patache is a fog oasis located in the Tarapacá Region of northern Chile, near the province of Iquique. This area is renowned for its unique fog collection techniques, used to study and develop water resource management strategies in arid and semi-arid regions. Alto Patache’s proximity to the coastline, at an altitude of approximately 780 meters, makes it a significant site for studying fog water collection and utilization. The study evaluated the impact of changes in fog water collection efficiency under different environmental conditions on local greenhouse vegetable production. It specifically estimated the energy requirements for regulating the greenhouse environment at both locations, including the energy needed for fans and water pumps. These energy needs were matched with the energy output from photovoltaic systems to ensure that the greenhouse environment remained within an optimal range (Albornoz 2023).

Some studies have shown the connection between environmentally related textile designs and electronic engineering. In their research, “Electrostatically driven fog collection using space charge injection,” Maher Damak and Kripa K. Varanasi proposed an innovative method to enhance fog collection efficiency through charge injection. Traditional fog nets often use mesh structures that rely on the inertial collision of fog droplets to capture them. However, these designs are limited by aerodynamics and have low efficiency, typically only 1% to 2%. By introducing an electric field, these aerodynamic limitations can be overcome.

The specific method involves using an ion emitter to charge the fog droplets, giving them a net charge, and then using an electric field to guide these charged droplets to the fog net. This approach overcomes air resistance and significantly improves fog collection efficiency. The low efficiency of traditional mesh designs is due to fog droplets being diverted by airflow. The electric field effectively guides the charged droplets to the mesh lines, independent of the airflow direction. Under the influence of the electric field, the movement trajectory of the fog droplets is redefined, with the field lines directing the droplets to the mesh lines, thereby enhancing collection efficiency.

This research demonstrates that by introducing an electric field into fog collection and using charge injection to alter the droplets’ movement trajectory, collection efficiency can be significantly improved. This method breaks through the limitations of traditional aerodynamics, providing an innovative solution and showing how combining electric fields with textile structures can improve fog collection technology (Figure 5).

It is worth noting that there are already many precedents for combining this fog collection technology with architectural or wearable art installations. Examples include the Warka Water project in southern Ethiopia, the CloudFisher project in Morocco, and Pavels Hedström’s art piece Fog-X. These projects often draw inspiration from biomimetic designs based on the fog collection behaviors of species such as the spider and the Namib Desert beetle.

3.3 DSSC based perceptron

Another previous project of the H.Om.E Project, “The Mind of A Greenhouse,” inspired the idea of DIY solar cells and their integration into textiles. Zagesi Elementary School is a small school located at an altitude of 3,700 meters, characterized by extremely intense sunlight and abandoned solar panels scattered around. Utilizing local plant dyes to create inexpensive dye-sensitized solar cell (DSSC) glass seemed like an excellent educational material (Figure 6). In the first phase, the project explored how to produce these solar technologies associated with local plants in a small laboratory with limited costs and without professional equipment. This explained how the choice of artistic materials and technologies in the research was influenced by the community’s characteristics.

Unfortunately, due to the costs of fieldwork and personnel issues, we decided to suspend field investigations connected to the community environment in the second phase of the project. However, we continued material research, such as how to DIY large DSSCs in a small laboratory and other material experiments to optimize solar cell production processes. These experiments focused on solar-related production technologies without glass substrates, such as patterning photoelectrodes using screen printing technology and creating textile versions of photovoltaic cells on volcanic mineral fiber substrates. These experiments became crucial components of the artifacts introduced in this paper.

Using dye-sensitized solar cell (DSSC) technology as a conceptual carrier in the project serves three functions. Firstly, this technology has positive implications for the project community on both practical and educational levels. Secondly, its structural properties relate to artificial intelligence concepts: the semi-transparency of DSSC makes it promising for optically mimicking the multi-layer structure of deep convolutional neural networks (DCNNs). Thirdly, the primary components, processes, and structures of DSSC are similar to the main electronic components of next-generation artificial computation — memristors. This research path and accumulated knowledge can keep our artistic work well-connected with the future field of artificial intelligence.

Discussing artificial intelligence through the lens of the perceptron has significant historical relevance. The single-layer perceptron, proposed by Rosenblatt in 1958, is one of the earliest and simplest neural network models. The perceptron can be considered one of the origins of connectionism — a computational model that simulates human cognitive processes by assuming that intelligent behavior arises from the interconnections and interactions between many simple units. Connectionism borrows its fundamental ideas from neurobiology, positing that intelligence originates from the connections and interactions between neurons. The perceptron was one of the first models to realize this concept.

Inspired by the principles of the perceptron, this paper proposes using DSSC as planar digital image sensors similar to CMOS, combined with a layer of microlens array (MLA) from light field cameras, to create a planar camera capable of intelligently recognizing the environment without using an objective lens. Light field cameras use microlens arrays (MLA) to capture the angular and positional information of light, which can be used in post-processing to reconstruct images with different focal points and depths of field. The role of the objective lens is to focus the light from objects onto the microlens array, forming a clear intermediate image. However, in the absence of an objective lens, the post-processing techniques need to be more refined and complex to compensate for the lack of initial focusing. Nonetheless, given the premise of producing artworks, this allows for more lenient technical requirements.

Light field cameras use microlens arrays (MLA) to capture the directional and positional information of light, forming four-dimensional light field data (4D Light Field data). A 4D Light Field is a high-dimensional data representation that describes how light propagates through space, recording the direction and intensity of light at each position. The concept of the light field was first introduced by physicist Michael Faraday and later applied to computational photography and optical imaging technologies. These data can be used in post-processing to reconstruct images with different focal points and depths of field. Each microlens in the microlens array splits and focuses the incoming light onto different positions of the sensor, capturing light information from various angles. This allows the light field camera to flexibly adjust focus and depth of field in post-processing.

This proposal is supported by previous research. In the study by Chou et al., it was suggested that equipping microlens arrays (MLA) on dye-sensitized solar cells (DSSC) can enhance light capture efficiency. MLA can focus more light onto the active material, improving photoelectric conversion efficiency. Specifically, MLA helps reduce light reflection losses, increasing the amount of light entering the cell, thereby enhancing the overall performance of the cell (Chou 2014).

Due to the photoelectric conversion properties of DSSC, this environmental data can be converted into sound. Here, we will briefly describe our setup. We connected a 3 x 3 cm DSSC to a pre-trained RAVE (Real-time Audio Variational Auto-Encoder) model in the Max/MSP environment, played through the nn~ wrapper. This setup converts the laser patterns projected onto the DSSC into sound in real time. The experimental video demonstrates and compares the sound results with and without the RAVE model. This experiment observed the relationship between the projected images, the screen-printed titanium dioxide photoelectrode patterns, and the sound. In other words, the pattern of the photoelectrode can predetermine the result of the light received by the DSSC layer.

Since dye-sensitized solar cells (DSSC) can be patterned through screen printing technology, it means we can create non-uniform photosensitive layers. This property is utilized to produce compressed representations in the latent space of a variational autoencoder. For example, rully3.ts is a model pre-trained on Google Colab using the voice of Rully Shabara Herman, the lead singer of Senyawa. In the above figure, the rully3 model in the Max environment generates 16 streams of compressed representations. We can create 16 DSSCs with different patterns to generate streams that replace those from the encoder object and input them into the decoder object.

The ability of DSSC to be patterned and its relationship with the RAVE model’s sound, as well as how the transparency of DSSC can participate in the functioning of deep convolutional neural networks (DCNN) or variational autoencoders (VAE), warrants further in-depth research. However, the current study has already provided many valuable insights. For example, glass-based DSSCs may integrate well with generative architectural design due to their cuttability, which meets the needs of parametric design. These buildings can also be well designed based on the semi-transparency of DSSC, whether considering optical computation for artistic sensor artifacts or designing greenhouse buildings for habitation.

It is also worth mentioning the environmental design relevance between this technology and fog net designs, which could potentially be integrated into a comprehensive design in the future. For instance, the prior work “A(g)ntense” by one of the study’s participants, Satoru Sugihara, provides a wonderful preview and imagination of this integration.

Figure 11. A(g)ntense, work of Satoru Sugihara, was built and exhibited at the design exhibition Blindspot Initiative Group Exhibition in Los Angeles (2014).

3.4 Similarities Between Memristor and DSSC Textiles

At the final stage of the H.Om.E Project, I conducted several experiments on fabricating DSSC on textile substrates. Due to the high-temperature resistance of glass, sintering the coated titanium dioxide nano-layer at 500 to 550 degrees Celsius on glass substrates is a common and viable approach. In order to get rid of the glass substrates and conduct sintering on other materials, there are commonly two options: first, employ other high-temperature alternative substrates, and second, apply low-temperature sintering methods. These experimental processes are documented on the Hackteria wiki. Here, I will omit the detailed introduction of these experiments and instead discuss the significance of fabricating memristors and their feasibility as abstract transformation model devices in this project.

Figure 10. A woven DSSC prototype with volcanic mineral fiber substrate, demonstrating the possibility of integrating dye-sensitized solar cells into textiles. The volcanic woven mineral fibers were provided by FILAVA.

First, memristors are a highly promising future product with significant commercial potential, but they have not yet become widespread in the market. A memristor is an electronic component based on variable resistance, whose resistance can change according to the amount of historical current passing through it. This component was proposed by Leon Chua in 1971 and is considered the fourth fundamental circuit element, alongside the resistor, capacitor, and inductor. Memristors have a memory effect, maintaining their resistance state even when the power is turned off, making them widely applicable in non-volatile memory, neuromorphic networks, and synaptic simulations.

Memristors can perform analog computations, which differ from traditional digital computations. They can directly execute operations such as addition and multiplication, which are particularly important in neural networks and machine learning. The simple structure of memristors allows them to be easily integrated into existing semiconductor manufacturing processes and offers excellent scalability. There is also a profound mimicking relationship between memristors and brain neurons, making memristors ideal components for neuromorphic computing.

The combination of crossbar array architecture and matrix multiplication in memristor technology has brought revolutionary changes to the fields of computation and storage. The crossbar array architecture is a highly dense grid structure composed of horizontally and vertically arranged wires, with variable resistance elements (such as memristors) placed at each intersection. This structure allows precise adjustment of resistance values at each intersection for efficient data storage and computation. In terms of computation, the crossbar array architecture is particularly suitable for implementing matrix multiplication operations. Matrix multiplication is a fundamental operation in linear algebra, involving the multiplication and summation of elements from two matrices. The crossbar array architecture achieves parallel processing by applying voltage between rows and columns. The resistance value of each memristor intersection can represent an element in a matrix, and by simultaneously operating multiple intersections, the crossbar array can quickly complete large-scale multiplication and addition operations, which are crucial for applications requiring extensive matrix computations, such as neural network training and inference.

The weaving craft itself is a complex grid structure, which has a certain degree of similarity to the crossbar array. Combining memristors with weaving crafts faces technical and practical challenges. We need to develop sufficiently small and durable memristor components to suit the characteristics of woven structures. Additionally, designing and manufacturing these smart textiles require new technologies and processes to ensure their functionality and durability. Since the core components of both memristors and DSSCs are titanium dioxide-based nanomaterials, we have already acquired sufficient knowledge about the properties of nano-titanium dioxide in this study. This knowledge aids the research team in the next phase of developing memristor artifacts and improves our coordination ability in communicating with semiconductor experts.

4. Conclusion

This paper documents the process of conceptual transformation and possible implementation methods for converting abstract transformation models into artifacts within the H.Om.E Project. These methods consider and assess the feasibility difficulties and the strength of conceptual and material connections between artifact design and abstract transformation models. In fact, materializing abstract transformation models is akin to materializing deep learning tools, as abstract transformation models themselves are AI models. This study not only focuses on the relationship between textiles and AI but also explores the possibility of using textile structures as a language to express geographical locations. This indirectly preserves the potential to express Ingold’s concept of the network of things in the context of textile design. These design ideas also consider the connection with the communities involved in the original project. Beyond their artistic value, these designs also hold potential benefits for the community in a functionalist sense.

5. References

“Lightfield Camera.” n.d. Accessed July 28, 2024. https://cameramaker.se/Lightfield.htm.

Chou, Chun-Chi, Kuan-Yi Tsao, Chih-Chung Wu, Hongta Yang, and Chih-Ming Chen. 2015. “Improved Power Conversion Efficiency for Dye-Sensitized Solar Cells Using a Subwavelength-Structured Antireflective Coating.” Applied Surface Science 328 (February):198–204. https://doi.org/10.1016/j.apsusc.2014.12.021.

Damak, Maher, and Kripa K. Varanasi. 2018. “Electrostatically Driven Fog Collection Using Space Charge Injection.” Science Advances 4 (6): eaao5323. https://doi.org/10.1126/sciadv.aao5323.

Du, Ke-Lin, Chi-Sing Leung, Wai Ho Mow, and M. N. S. Swamy. 2022. “Perceptron: Learning, Generalization, Model Selection, Fault Tolerance, and Role in the Deep Learning Era.” Mathematics 10 (24): 4730. https://doi.org/10.3390/math10244730.

Fernandez, Daniel M., Alicia Torregrosa, Peter S. Weiss-Penzias, Bong June Zhang, Deckard Sorensen, Robert E. Cohen, Gareth H. McKinley, Justin Kleingartner, Andrew Oliphant, and Matthew Bowman. 2018. “Fog Water Collection Effectiveness: Mesh Intercomparisons.” Aerosol and Air Quality Research 18 (1): 270–83. https://doi.org/10.4209/aaqr.2017.01.0040.

Junyi Li, Fulin Peng, Fan Yang, Xuan Zeng. 2024. “A Memristor Crossbar-Based Computation Scheme with High Precision.” n.d. Ar5iv. https://ar5iv.labs.arxiv.org/html/1611.03264.

Ki, Ravikumar, and Sukumar R. 2022. “Memristor Based Object Detection Using Neural Network.” High-Confidence Computing 2 (4): 100085. https://doi.org/10.1016/j.hcc.2022.100085.

Mariani, Paolo, Antonio Agresti, Luigi Vesce, Sara Pescetelli, Alessandro Lorenzo Palma, Flavia Tomarchio, Panagiotis Karagiannidis, Andrea C. Ferrari, and Aldo Di Carlo. 2021. “Graphene-Based Interconnects for Stable Dye-Sensitized Solar Modules.” ACS Applied Energy Materials 4 (1): 98–110. https://doi.org/10.1021/acsaem.0c01960.

Martineau, David. n.d. “Dye Solar Cells for Real.”

Rosner, Daniela K. 2018. Critical Fabulations: Reworking the Methods and Margins of Design. MIT Press.

Wei, Tzu‐Chien, Jo‐Lin Lan, Chi‐Chao Wan, Wen‐Chi Hsu, and Ya‐Huei Chang. 2013. “Fabrication of Grid Type Dye Sensitized Solar Modules with 7% Conversion Efficiency by Utilizing Commercially Available Materials.” Progress in Photovoltaics: Research and Applications 21 (8): 1625–33. https://doi.org/10.1002/pip.2252.

Xu, Weilin, Jingjuan Wang, and Xiaobing Yan. 2021. “Advances in Memristor-Based Neural Networks.” Frontiers in Nanotechnology 3 (March). https://doi.org/10.3389/fnano.2021.645995.

Zhu, Shuaishuai, Andy Lai, Katherine Eaton, Peng Jin, and Liang Gao. 2018. “On the Fundamental Comparison between Unfocused and Focused Light Field Cameras.” Applied Optics 57 (1): A1. https://doi.org/10.1364/AO.57.0000A1.