Neuroscience Researcher Calls Neuralink Outdated, Discusses PB-Scale Storage and Optical Interconnects for Brain-Computer Interfaces
English summary
A neuroscience researcher described Neuralink as scientifically opportunistic and behind academic standards, favoring engineering over rigorous science. She detailed that their lab builds custom 22nm low-power chips for implantable neural recording, requiring up to 10 PB of storage to capture 30 kHz signals from hundreds of electrodes—raw data at 80 GB per probe per hour. The cost of maintaining PB-scale storage can reach millions of dollars over five years, forcing labs to buy only 1 PB at a time. She noted that Harvard and Google's 1 mm³ human brain dataset is 1.4 PB, while a whole mouse brain would be ~1 EB. The conversation also covered the shift from copper to optical interconnects, echoing NVIDIA's recent silicon photonics switch for AI data centers, as copper cannot sustain 1.6 Tb/s over rack-scale distances.
Chinese summary
一位神经医学研究者称，在学术界看来，马斯克的Neuralink技术落后，属于“工程领先、科学取巧”。她所在实验室自行设计22纳米低功耗芯片用于植入式神经记录，因需以30 kHz频率采集数百个电极信号，每小时单个探针产生80 GB数据，导致总存储需求达10 PB。维护PB级存储五年成本可达百万美元，研究经费只能每次购买1 PB。她举例哈佛与谷歌1 mm³人脑全分辨率数据为1.4 PB，整个鼠脑估计需1 EB。此外，讨论了铜互连向光互连的转变，呼应了英伟达为百万GPU数据中心发布的硅光子交换机，因铜在1.6 Tb/s下传输距离不足机柜高度。
Key points
Neuralink is viewed by academic researchers as technologically backwards, prioritizing press-friendly engineering over rigorous science.
学术圈认为Neuralink技术落后，重公关式工程而轻严谨科学。
Lab designs custom 22nm chips for implantable neural recording, operating at 0.5V, ideal for low-power bio-signal acquisition.
实验室自行设计22纳米低功耗模拟芯片，在0.5V电压下工作，专用于植入式神经信号记录。
Neural signal sampling at 30 kHz with hundreds of electrodes generates 80 GB/hour per probe, requiring up to 10 PB total storage.
以30 kHz频率对数百个电极采样，单个探针每小时产生80 GB数据，总存储需求达10 PB。
PB-scale storage maintenance costs ~$1 million over five years, forcing labs to purchase one petabyte at a time; a whole mouse brain would need about 1 exabyte.
PB级存储五年维护成本约百万美元，实验室只能每次买1 PB；完整鼠脑数据量估计需1 EB。
Transition from copper to optical interconnects is driven by copper's inability to sustain 1.6 Tb/s over data-center rack heights, aligning with NVIDIA's CPO silicon photonics switch announcement.
铜在1.6 Tb/s速率下传输距离不足机柜高度，驱动互连从铜向光转变，与英伟达共封装光学交换机发布趋势一致。