The remarkable human brain that conceived heart surgery and interplanetary travel runs on roughly the same amount of energy as a refrigerator bulb, thanks to its trillions of synapses. Scientists are now trying to replicate this ability with printable, flexible, biocompatible, and recyclable artificial synapses that don't require the rigid metals and complex manufacturing processes used in existing artificial synapse components.
Researchers at Suzhou University of Science and Technology in China have developed a flexible array of 100 artificial electronic synapses that can switch electrical states in just 26 nanoseconds and be chemically broken down within six minutes.
The printable components also showed brain-like learning and memory, achieving 97.23% accuracy in a simulated handwriting recognition task. The researchers believe the technology could pave the way for more efficient, affordable, and environmentally friendly computing hardware.
One of the reasons behind the brain's remarkable efficiency is that memory and processing take place in the same physical real estate. Modern computers, by contrast, typically have separate memory and processing units. Traditional CPUs excel at sequential computing tasks, such as calculating a massive spreadsheet or running a precise physics simulation, as these often require executing steps in strict order at blazing speeds.
However, the complex parallel processing that applications such as artificial intelligence demand is a different ballgame altogether. AI involves processing billions of interconnected variables, shuffling huge amounts of data back and forth between memory and processing units. Even with specialized parallel-processing hardware, this consumes immense amounts of energy, and creates a digital traffic jam that can severely limit performance.
To address these bottlenecks, engineers have tried to replicate the brain’s unified structure within computing frameworks using rigid artificial synapses, including two-terminal memristors and synaptic transistors. These nanoscale components are usually made from rigid silicon and transition metal oxides, such as hafnium oxide (HfOx) or titanium dioxide (TiO2), and layered alongside dense, heavy metals like platinum or tungsten.
However, they come with several drawbacks. For starters, the microscopic metallic filaments inside the metal-oxide layers grow in highly unpredictable, chaotic paths, making it incredibly difficult for engineers to achieve uniform, reliable performance across millions of artificial synapses on a single chip.
Furthermore, many conventional devices use brittle, rigid materials that cannot easily stretch or conform to irregular surfaces, limiting their suitability for soft robotics and seamless biological integration. Some material combinations also raise biocompatibility concerns, particularly for medical implants. In addition, many of these chips require complex manufacturing processes, including energy-intensive vacuum deposition and semiconductor fabrication facilities.
To address these issues, the researchers developed an artificial electronic synapse using a hydrogen-bonded organic framework (HOF), a highly organized network of organic molecules held together by hydrogen bonds. These bonds act as tiny structural supports, keeping the molecules arranged in a stable, orderly pattern.
The researchers created their framework, called TCPP-HOF, using organic molecules known as porphyrins that assemble into flat, two-dimensional sheets stacked approximately 0.35 nanometers apart, creating an orderly structure that allows electrical charges to move through the material relatively easily. This arrangement addresses a longstanding problem with organic electronic materials, which tend to suffer from poor electrical conductivity and unstable charge transport.
When a positive voltage is applied, electrical charges move through the molecular structure, with electrons becoming trapped at specific sites within the framework. This creates a more conductive pathway, reducing the material's electrical resistance. A negative voltage reverses the process, releasing the trapped charges and increasing resistance again. By controlling these changes, the device can mimic how biological synapses strengthen or weaken their connections as we learn and form memories.
To build the device, the researchers sandwiched the TCPP-HOF layer between two conductive MXene electrodes. They then deposited the arrangement onto a flexible polyimine plastic substrate, producing a 10 x 10 array containing 100 artificial synapses. Interestingly, the electrodes and active material were deposited using liquid-based coating techniques, eliminating the need for the expensive vacuum-deposition equipment commonly used to manufacture conventional semiconductor devices.
The team then subjected the synapses to a series of electrical tests, demonstrating their ability to imitate several biological learning mechanisms, including long-term potentiation and depression, where repeated stimulation strengthens or weakens synaptic connections. The devices could reliably maintain 10 distinct electrical conductance levels, allowing them to represent different connection strengths. They also achieved an electrical switching time of just 26 nanoseconds, making them particularly fast compared with previously reported HOF-based artificial synapses.
To assess their computing potential, the researchers used the devices’ measured electrical characteristics in a convolutional neural network designed to recognize handwritten digits from the standard MNIST dataset. The network achieved 97.23% recognition accuracy, approaching the 99.24% software reference result. However, the recognition task was evaluated using a network model representing approximately 120,000 synapses, rather than running entirely on the physical 100-device array.
The researchers also tested the device's flexibility by bending it 200 times consecutively around a radius of about 5 mm (0.2 inches). Its electrical behavior remained largely unchanged, and the handwriting recognition accuracy after bending still reached 96.33%. This flexibility, combined with the relatively simple manufacturing process, could make the technology particularly attractive for wearable health monitors, electronic skin, soft robotics, and other applications where conventional rigid electronics are difficult to integrate.
Then there's the disappearing act. The researchers designed the polyimine substrate to break down when immersed in a heated chemical solution containing specific amine compounds. Within six minutes, the polymer decomposed into soluble constituents, leaving only residual material behind. This offers a potential route toward recyclable electronics and temporary devices that can be intentionally dismantled after use, reducing electronic waste and potentially providing a means of physically destroying sensitive information.
However, the technology remains at an early experimental stage. The researchers have demonstrated impressive switching speed and promising neural network performance, but have yet to establish how the devices would perform in large-scale, fully integrated computing systems. Questions also remain about their long-term durability, energy consumption, manufacturing consistency, and suitability for direct contact with living tissue. The chemical decomposition process also requires a controlled treatment.
According to the researchers, the next steps involve expanding the 100-device prototype into larger arrays and developing protective barriers to prevent environmental moisture from interfering with its operation. If they can overcome these challenges, the technology could eventually enable affordable, flexible, brain-inspired electronics that process information directly where it is generated and can be dismantled when their useful life is over.
The research was published in the International Journal of Extreme Manufacturing.
Source: EurekAlert
