- Blog
- Computing
Axonally -Connected Brain Organoids: PioneeringBringing Innovation into Next-Generation Computing — Toward the Realization of the Brain Processing Unit (BPU) —
#Others
Jan 17, 2025
SoftBank Corp.


Blogs
1. Successful Information Processing Using Brain Organoids Connected via Axonal Networks
SoftBank Corp. and The University of Tokyo have conducted joint research on next-generation computing technology using cerebral organoids.*1 As a result, we demonstrated that axon-connected cerebral organoids can process information with higher precision compared to individual cerebral organoids. This finding highlights the importance of circuitry in the functionality of cerebral organoids and suggests the potential for future applications in next-generation computing.
The results of this research were presented at "Neuroscience 2024" (Society for Neuroscience), held in Chicago, USA, from October 5 to 9, 2024. We plan to publish these findings in a scientific journal in the near future.
*1 Cerebral organoids are three-dimensional tissue culture models that partially replicate the structure of brain tissue, created from human pluripotent stem cells (iPS cells or ES cells).
*2 Axons are long, slender projections that extend from neurons (nerve cells). They transmit electrical signals to other nerve cells or various types of cells.
2. Background of the Joint Research
With the proliferation of generative AI, the power consumption of data centers is rapidly increasing. For instance, training GPT-4 requires over 50 gigawatt-hours (GWh) of electricity, which is equivalent to the monthly power consumption of approximately 55,600 average American households. In contrast, the human brain operates on a mere 20 watts and can perform a wide variety of complex tasks. Additionally, the amount of data required for learning new tasks is significantly lower in the human brain compared to AI, highlighting the brain's high learning efficiency and adaptability.
SoftBank’s Research Institute of Advanced Technology is focusing on these characteristics of the brain to conduct research on our application to computing. The technology of interest is the small artificial brain tissue known as "cerebral organoids," created using pluripotent stem cells. By controlling cerebral organoids through electrical stimulation, it is considered possible to develop accelerators called "Brain Processing Unit (BPU)" that leverage the strengths of the brain. Unlike traditional silicon-based accelerators, these brain-inspired accelerators use the neurons of cerebral organoids, which are organic material, to process electrical signals and perform computations, constituting an entirely new computing technology.
Several previous studies have reported success in making cultured neural tissues, such as cerebral organoids, perform tasks like simple games and classification. However, the accuracy of tasks in these studies has generally been low. One contributing factor is that the cultured neural tissues used in previous research are much smaller in size (typically a few millimeters) and have immature neural circuits compared to the human brain. Going forward, in order to advance applications in computing, it is necessary to enhance the sophistication of neural circuits, which are the foundation of the brain's task processing capabilities.
Against this background, SoftBank and The University of Tokyo have undertaken research aimed at improving the accuracy of information processing by connecting cerebral organoids via axons to replicate a structure more closely resembling the human brain.

Figure1. Overview of the Brain Processing Unit (BPU)
3. Details and Results of the Joint Research
In this study, three groups were prepared: a single cerebral organoid ("Solo"), two cerebral organoids connected via axons ("Duo"), and three cerebral organoids connected via axons ("Trio"). Each group was subjected to continuous stimulation with two different spatiotemporal patterns. The activity data of the cerebral organoids immediately after stimulation were then classified using machine learning algorithms, and the accuracy of stimulus classification was compared among Solo, Duo, and Trio.

Figure2. Overview of the Experiment
Figures 3 and 4 illustrate the time course of classification accuracy using SVM (Support Vector Machine) and 2D CNN (two-dimensional Convolutional Neural Network). For both algorithms, it was observed that the classification accuracy of Trio improved over time. These results suggest that cerebral organoids connected via axons may have a higher capability for precise information processing compared to single cerebral organoids.

Figure3. Time Course of Classification Accuracy Using SVM

Figure4. Time Course of Classification Accuracy Using 2D CNN
Figure 5 shows graphs comparing the activity patterns of cerebral organoids before and after training. Prior to training, the differences in activity patterns immediately following each stimulus were small. However, after training, these differences became more pronounced. This indicates that the neural circuits and responses of the cerebral organoids were altered by continuous stimulation, contributing to the improvement in classification accuracy.

Figure5. Activity Patterns of Cerebral Organoids Before and After Training
(Top) Example of Activity Data Immediately Following Each Stimulus Before and After Training
(Bottom) Normalized Heatmap Showing Differences in Activity Immediately Following Each Stimulus Before and After Training
4. Future Prospects
This joint research has demonstrated that cerebral organoids connected via axons can process information with higher accuracy compared to single cerebral organoids. This suggests that the information processing capabilities of cultured neural tissues may be significantly enhanced with further advancements in cultivation techniques. These findings represent an important milestone, indicating the potential for future applications in computing.
In the near future, we are planning to hold events to widely disseminate the future vision of applying cerebral organoids to computing, based on the methods and insights obtained from this joint research. For more details, please visit our dedicated website.
Brain Processing Unit - 生命とコンピューターが融合する未来 -(Japanese page only)