Found another possible use for Pico.
Valve CEO to release Neuralink competitor brain chip by year-end
Valve co-founder and CEO **e Newell's neural interface startup, Starfish Neuroscience, has announced plans to release its first brain chip in late 2025, positioning itself as a competitor to Elon Musk's Neuralink with a focus on creating smaller, less invasive implants capable of simultaneously accessing multiple brain regions.
Published
May 25, 2025
Starfish Neuroscience Origins
Starfish Neuroscience was co-founded in 2019 by **e Newell and Philip Sabes, a former co-founder of Neuralink and professor emeritus at UCSF
12. The company is based in Bellevue, Washington—the same city as Valve's headquarters—reflecting Newell's long-standing interest in brain-computer interfaces that began over a decade ago when Valve employed in-house psychologists to study biological responses to video games
12. This interest became public when Valve explored brain-computer interface concepts at the Game Developers Conference in 2019
1.
The company's name draws interesting parallels with scientific research on actual starfish, which have revealed insights into neural function evolution. Biologists have discovered 40 neuropeptide genes in starfish that are similar to chemicals found in human brains, demonstrating how neural signaling has evolved over hundreds of millions of years
34. This connection between starfish neuroscience research and the company's focus on neural interfaces suggests an approach rooted in evolutionary understanding of brain function.
Ultra-Low Power Neural Chip
Ultra-low power neural chips represent a critical advancement in brain-computer interface technology, enabling AI processing at the edge with minimal energy consumption.
These specialized processors, like BrainChip's Akida Pico, can operate on less than 1 milliwatt of power while supporting neural network models for applications such as voice detection and audio enhancement12. This efficiency is crucial for implantable neural interfaces that must function within the body's limited energy environment.The field has seen remarkable innovations from multiple companies. Korean researchers developed the C-Transformer chip that reportedly uses 625 times less power than Nvidia's A100 GPU while handling large language model processing
3. Other notable developments include ROHM's on-device learning AI chip consuming just tens of milliwatts (1000× less than conventional learning-capable AI chips)
4, and ultra-efficient chips from Chinese scientists that consume less than 2 microjoules per instance while maintaining over 95% accuracy in speech recognition
5. These advancements in neuromorphic computing—which mimics the brain's neural structure—create possibilities for neural interfaces that can process information locally without excessive heat generation or battery drain.
Multiple Brain Region Access
Multi-region brain interfaces represent a significant advancement in neural technology, allowing simultaneous monitoring and stimulation across different brain areas. Unlike single-region interfaces, these systems can record neural activity from both surface and deep brain structures concurrently, providing unprecedented insights into how different parts of the brain communicate during learning, memory formation, and other cognitive processes.
1Recent research demonstrates the clinical potential of multi-region interfaces. Scientists have developed systems that simultaneously monitor the primary somatosensory cortex (S1) and anterior cingulate cortex (ACC) to detect pain signals, while delivering therapeutic stimulation to the prefrontal cortex (PFC).
2This approach has proven effective for both acute and chronic pain in animal models. The technology's versatility extends beyond pain management, with applications including motor brain-machine interfaces for paralyzed patients using Utah arrays to transmit neural signals from motor cortex regions to external devices.
3With thousands of electrodes distributed across multiple brain areas, as demonstrated by Neuralink's flexible electrode "threads" system capable of deploying 3,072 electrodes across 96 threads, these interfaces promise to revolutionize our understanding of distributed neural activity and enable more sophisticated brain-computer interactions.
(20min delay)