As generative AI systems advance, their energy demands are growing. Training and running large language models consume vast amounts of electricity. 

It's estimated that AI's energy needs will double over the next five years, accounting for 3% of global electricity consumption. But what if AI chips could operate more like the human brain, handling complex tasks with minimal energy consumption? A growing number of scientists and engineers believe that organoid intelligence may hold the key.

This past July, at the United Nations "AI for Good" summit in Geneva, AI enthusiasts were first introduced to the concept of "brain-inspired chips." David Gracias, a professor of chemical and biomolecular engineering at Johns Hopkins University, delivered a speech at the conference, discussing his latest research on biochips and their applications in AI. Gracias's research team, whose expertise focuses on nanotechnology, intelligent systems, and bioengineering, is one of the first to build a functional biochip. This chip combines neural organoids with advanced hardware, enabling the chip to operate on and interact with living tissue.

Organoid intelligence is an emerging field that combines lab-grown neurons with machine learning to create a new form of computing. (The term "organoid intelligence" was coined by Johns Hopkins researchers, including Thomas Hartung.) These neurons are called organoids, and more specifically, they are three-dimensional clusters of lab-grown brain cells that can mimic neural structure and function. Some researchers believe that so-called biochips - organoid systems that integrate living brain cells into hardware - have the potential to surpass silicon-based processors like CPUs and GPUs in efficiency and adaptability. Experts say that if commercialized, biochips could potentially reduce the staggering energy demands of today's AI systems while enhancing their learning capabilities.

As generative AI systems advance, their energy needs are growing. Training and running large language models consume vast amounts of electricity. AI's energy needs are projected to double over the next five years, accounting for 3% of global electricity consumption. But what if AI chips could function more like the human brain, handling complex tasks with minimal energy? A growing number of scientists and engineers believe the key may lie in organoid intelligence.

In July of this year, at the United Nations "AI for Good" Summit in Geneva, AI enthusiasts were first introduced to the concept of "brain-inspired chips." David Gracias, Professor of Chemical and Biomolecular Engineering at Johns Hopkins University, delivered a speech at the event, discussing his latest research on biochips and their applications in AI. Gracias's research team, specializing in nanotechnology, intelligent systems, and bioengineering, is one of the first to build a functional biochip. The chip combines neural organoids with advanced hardware, enabling the chip to operate on and interact with living tissue. Organoid intelligence is an emerging field that combines lab-grown neurons with machine learning to create a completely new form of computing. (The term "organoid intelligence" was coined by researchers at Johns Hopkins University, including Thomas Hartung.) These neurons, called organoids, are more specifically three-dimensional clusters of lab-grown brain cells that can mimic neural structure and function. Some researchers believe that so-called biochips - organoid systems that integrate living brain cells into hardware - have the potential to surpass silicon-based processors like CPUs and GPUs in efficiency and adaptability. Experts say that if biochips are commercialized, they could potentially reduce the staggering energy demands of today's artificial intelligence systems while enhancing their learning capabilities.

"This is an exploration of a different way to form computers," Gracias said.

How do biochips simulate the brain?

Traditional chips have long been confined to two-dimensional layouts, which restrict how signals are transmitted within the system. This paradigm is beginning to shift as chipmakers are developing 3D chip architectures to boost the processing power of their devices.

Similarly, biochips are designed to mimic the three-dimensional structure of the brain itself. The human brain can support up to 200,000 connected neurons—a level of interconnectivity that is unattainable with planar silicon chips, according to Gracias. This spatial complexity enables biochips to transmit signals across multiple axes, enabling more efficient information processing.

Gracias's team has developed a 3D electroencephalography (EEG) shell that wraps around organoids, enabling richer stimulation and recording than traditional planar electrodes. The shell conforms to the organoid's curved surface, creating a better interface for stimulating and recording electrical activity.

To train the organoids, the team employed reinforcement learning. Electrical pulses are applied to targeted areas. When the resulting neural activity matches the expected pattern, it is reinforced by the brain's natural reward chemical, dopamine. Over time, the organoids learn to associate certain stimuli with outcomes. Once a pattern is learned, it can be used to control physical actions, such as steering a microrobot car, using strategically placed electrodes. This demonstrates neuromodulation—the ability of organoids to generate predictable responses. These consistent responses lay the foundation for higher-level functions, such as stimulus discrimination, which is crucial for applications such as facial recognition, decision-making, and generalized AI reasoning.

Gracias's team is in the early stages of developing a microscopic autonomous car controlled by a biochip: this demonstrates that the system can function as a controller. This experimental work foreshadows future applications in robotics, prosthetics, and bio-integrated implants that communicate with human tissue.

These systems also hold great promise for disease modeling and drug testing. Gracias's group is developing organoids that mimic neurological disorders such as Parkinson's disease. By observing how these diseased tissues respond to various drugs, researchers can test new treatments in a petri dish, rather than relying solely on animal models. They can also uncover underlying mechanisms of cognitive impairment that current AI systems cannot simulate.

Because these chips are living, they require constant care: temperature regulation, nutrient supply, and waste removal. Gracias' team kept the integrated biochip alive and functioning properly for up to a month through continuous monitoring.

Challenges in Scaling Biochip Technology

However, significant challenges remain. Biochips are fragile and costly to maintain, and current systems rely on bulky laboratory equipment. Scaling them down to practical applications requires biocompatible materials and technologies that can autonomously manage life-support functions. Neural latency, signal noise, and the scalability of neuron training also pose obstacles to real-time AI inference.

"There are a lot of biological and hardware issues," Gracias said.

Meanwhile, some companies are testing the waters. Swiss startup FinalSpark claims its biochip can store data in living neurons - a milestone the company calls "biobits," according to Ewelina Kurtys, a scientist and strategic advisor at the company. This suggests that biological systems could one day perform core computing functions traditionally handled by silicon hardware, rivaling digital processors in performance while achieving exponentially greater energy savings. "The biggest challenge is programming the neurons, because we need to find a completely new approach," Kurtys said.

However, the transition from lab to industry requires more than just technological breakthroughs. "We have enough funding to keep the labs running," Gracias said. "But to get the research off the ground, we need more funding from Silicon Valley."

Whether biochips will augment or replace silicon chips remains to be seen. But as AI systems become increasingly power-hungry, the idea of chips that can think and consume energy like the brain is becoming increasingly attractive.

For Gracias, this technology could be coming to market sooner than we think. "I don't think there will be any major obstacles to implementation," he said.

Source: Content compiled from IEEE

Reference link: https://spectrum.ieee.org/biochip-organoid-intelligence-ai-processor?utm_source=homepage&utm_medium=hero&utm_campaign=2025-08-09&utm_content=hero2

View more at EASELINK