Automatic artificial synapses mimic human color perception

A new artificial synapse produces its own electricity that can distinguish colors with near-human precision, potentially revolutionizing how edge devices process visual information.

The device was developed by researchers at the Tokyo University of Science and achieved discrimination at 10 nanometer wavelength while consuming minimal power, two major obstacles to computer vision technology.

Unlike traditional machine vision systems, capturing every detail at 10-60 frames per second, this biologically inspired approach mimics how human synapses selectively filter visual information. The result is that power consumption is greatly reduced without sacrificing recognition capabilities.

Bipolar response enables complex logic operations

The most striking feature of the device is its bipolar voltage response at different wavelengths. Blue light produces a positive voltage, while red light produces a negative voltage, a feature that enables a single device to perform multiple logic operations simultaneously.

This wavelength-dependent polarity switching represents that it usually operates in a narrow voltage range compared to existing artificial synapses. Researchers demonstrated the sum or XOR logic functions using changes in light intensity and wavelength, which often require multiple conventional devices.

What makes this possible? The team integrates two different dye-sensitized solar cells, each responding to a different wavelength range. When two cells are illuminated simultaneously, their combined responses produce unique biphasic behavior.

Six-digit classification exceeds previous functions

Tests show that the device can distinguish input patterns of up to six-digits—significantly exceeding the four-digit classification that is typically achieved through conventional artificial synapses. This enhanced functionality stems from an unusually broad paired promotion index range of -3,776 to 8,075 compared to conventional devices in the 100-200 range.

The research team evaluated the classification performance of 4-bit to 7-bit operations. While single-ended solar cells struggle in the 5-bit classification, bipolar devices maintain clear separation through 6-bit operation, with minimal overlap between different input states.

A key finding that applications often overlook: when different wavelengths alternate, the device’s reset mechanism is faster than the natural relaxation time than the natural relaxation time. “Alternating light wavelengths between red and blue may facilitate VOUT resetting to 0 V faster, as blue and red light causes opposite polarities,” the researchers noted. The feature can operate continuously without waiting time, often limiting artificial synaptic performance.

Key Performance Indicators:

• 10nm wavelength discriminatory resolution
• Six-bit input mode classification
• The accuracy of multicolor motion recognition task is 82%
• Automatic operation to eliminate external energy demand

Motion recognition applications in the real world

To demonstrate practical application, the researchers tested the device on a multicolor motion recognition task involving six human actions recorded in red, green and blue. The system achieves an overall accuracy of 82% while maintaining perfect color discrimination.

The device responds differently to each color: the positive voltage of red light, the negative blue color and the green color is close to zero. This three-state response mode can replace multiple photodiodes currently required in traditional color video systems.

According to Associate Professor Takashi Ikuno, who leads the research, “We believe this technology will help achieve color discrimination capabilities close to the human eye and be used in optical sensors for autonomous vehicles, low-power biometric sensors, for medical purposes and portable recognition devices.”

Overcome energy limitations in edge computing

Current machine vision systems face a fundamental challenge: processing large amounts of visual data requires a lot of computing resources and power. This limitation affects battery life especially vital to edge devices such as smartphones, drones and autonomous vehicles.

The new device solves this problem by using automatic design of dye-sensitized solar cells. It does not require external voltages, such as artificial synapses based on photocurrent, but instead generates electricity through solar energy conversion while maintaining high sensitivity throughout the visible spectrum.

The study is based on the principle of physical reservoir computing, where material properties handle complex calculations rather than traditional digital processors. This approach reduces training costs by converting time series data through the intrinsic dynamics of the device, requiring only lightweight output processing.

Future applications across industries

These implications extend multiple departments. Self-driving cars can benefit from more effective recognition of traffic signals and obstacles. Healthcare applications may include wearable devices for monitoring blood oxygen levels with minimal battery capacity. Consumer electronic devices can see smartphones and AR/VR headsets, and their battery life can be significantly improved.

However, actual implementation faces certain challenges. The current logic level determination depends on specific voltage thresholds that may require other circuits to operate reliably under actual conditions.

The study, published in a scientific report, is an important step in bringing complex visual recognition into battery-powered devices. By mimicking how human vision can process information selectively, this technology can enable everyday devices to see the world more efficiently than ever before.