The present computer systems excel at precise calculations, but as the demand for AI-based applications grows, there is a need for more efficient systems capable of processing data in real-time while maintaining accuracy. Eveline van Doremaele, a researcher from TU/e, is tackling this challenge by working on the development of a novel generation of computers inspired by the human brain. Notably, she has utilized organic materials to create a unique chip that incorporates neuromorphic computing, enabling it to interact with our bodies.
AI-driven applications such as self-driving cars, facial recognition, and language recognition rely on computer systems that can adapt to dynamic environments and handle unstructured and imperfect data. While current artificial neural networks perform reasonably well, they also suffer from significant drawbacks such as high energy consumption and time-consuming complex calculations.
In response to these limitations, Eveline van Doremaele has dedicated her efforts over the past few years to designing a cutting-edge computer system. Her research has resulted in the development of a smart chip that holds promise for diverse applications within the human body. On Thursday, May 25, she successfully defended her thesis with distinction at the Mechanical Engineering department of TU/e.
Mimicking the brain
“The human brain, a remarkable system capable of handling complex tasks, holds the key to efficient problem-solving and adaptability,” remarked Van Doremaele, gently tapping her head. She emphasized that our brain’s ability to concurrently perform processes, calculations, and learn from past experiences makes it ideal for AI applications. Consequently, neuromorphic computing, which emulates the brain’s structure and functionality in computer systems, has gained significant traction in recent years. Van Doremaele stated, “Our brain serves as a remarkable model for energy-efficient, agile, and dynamic computing systems, inspiring scientists and our research group alike. We aim to take it a step further by developing a device that fosters self-learning interactions between humans and machines.”
She elaborated on potential applications, such as a smart prosthetic arm that can be connected to the human body and taught to manipulate objects using artificial neurons. Additionally, a chip incorporating multiple sensors simultaneously could detect circulating cancer cells amidst millions of normal cells, and an adaptive pacemaker could cater to the changing needs of an aging heart. Van Doremaele expressed excitement about the limitless possibilities once the technology is fully realized.
Self-learning system
In her quest to develop such a chip, Van Doremaele embarked on a search for suitable materials that possess both programmability and biocompatibility. Her research led her to discover the promising capabilities of conductive organic polymers. These elongated molecules facilitate the flow of electric current and prove highly effective in meeting the desired criteria.
To achieve the crucial aspect of self-learning in the system, Van Doremaele emphasized the importance of variable resistance within the device. She drew parallels to our brain, where connections between neural cells strengthen with repeated learning experiences. By utilizing ions, the resistance can be adjusted accordingly. However, the aim is to establish a permanent connection, enhancing the longevity of the learning process.
Van Doremaele’s exploration of conductive organic polymers and their potential in creating self-learning chips paves the way for exciting advancements in the field of neuromorphic computing.
Weaker connections
“In our field, it has been customary to employ materials where the connections gradually weaken over time,” remarked the Ph.D. candidate. She further explained that such a scenario would imply, for instance, forgetting how to grasp a pen after only a month with a prosthetic arm.
The tested material, P-3O, exhibits unique properties. As an ambipolar material, it possesses the ability to adjust resistance while maintaining the established connections. Notably, P-3O is compatible with both liquid electrolytes, such as those found within the watery environment of the human body, and solid electrolytes like ion gels. By interconnecting cells, intricate circuits can be created with specific characteristics. This becomes advantageous in situations involving the measurement of faint signals, like subtle muscle movements, or signals surrounded by significant noise, such as a heartbeat.
The introduction of P-3O as a viable material opens up exciting possibilities for the development of advanced systems capable of adapting, learning, and accurately capturing delicate or noisy signals in various applications.
Measuring sweat samples
While extensive research is still required for complex measurements, Van Doremaele has already leveraged neuromorphic computing to develop a biosensor capable of analyzing sweat samples for cystic fibrosis, a hereditary disease. The chip employs various sensors to measure the potassium and chlorine content in sweat. Van Doremaele conducted tests where the system made predictions for each sample, and upon incorrect predictions, she manually corrected the system by pressing a button. Ultimately, the biosensor achieved accurate results by learning in a unique manner reminiscent of neurons in the human brain. This achievement serves as a solid foundation for further advancements in the field.
Van Doremaele’s work has garnered considerable interest, given the widespread presence and future ubiquity of AI. The escalating energy demands of data centers pose a pressing challenge, necessitating the exploration of alternative computer systems. The focus on organic materials for self-learning biomedical applications sets her research apart as a distinctive endeavor.
Due to the multidisciplinary nature of the project, Van Doremaele has established collaborations within her university campus. By seeking colleagues with diverse backgrounds and fostering knowledge-sharing, she serves as a vital link between the research institutes EAISI (Artificial Intelligence) and ICMS (Complex Molecular Systems) at TU/e. While the journey of a Ph.D. candidate can be solitary at times, Van Doremaele takes pride in the significant recognition she has gained through her dissertation.