姓名：吴璐坤 学号：22021110295 学院：电子工程学院
原文链接：https://neuralink.com/approach/
【嵌牛导读】关于Neuralink公司的一些信息以及脑机接口的一些技术细节，我们在这篇文章中进行一些分析。信息的主要来源是Neuralink官网的信息。本文主要分析目前脑机接口所面临的主要挑战
【嵌牛鼻子】脑机接口 Neuralink AI
【嵌牛提问】WHAT ARE THE BIGGEST CHALLENGES IN MAKING A SCALABLE BCI?
【嵌牛正文】
Neuralink’s technology builds on decades of BCI research in academic labs, some of which is currently being tested in ongoing clinical studies. The BCI systems used in these aforementioned studies have no more than a few hundred electrodes with connectors that pass through the skin. Their use also requires laboratory equipment and personnel to be present. Our challenge is to build a safe and effective BCI that is wireless and fully implanted, scales up the number of electrodes, removes the need for external equipment (other than the device being controlled), and that users can take anywhere and operate by themselves. Recent engineering advances in the field and new technologies developed at Neuralink are paving the way for progress on each of these key technical hurdles.
主要的目标是构建安全高效无线的脑机接口，提升电极数量的规模，避免外接设备的引入。
ELECTRODES 高分辨率耐腐蚀电极
In order to optimize the compatibility of our threads with the surrounding tissue, we believe that they should be on the same size scale as neighboring neurons and as flexible as possible. The threads also have to resist corrosion from fluid in the tissue. Therefore, we microfabricate the threads out of thin film metals and polymers. To meet these criteria, we’ve developed new microfabrication processes and made advances in materials science. These include the integration of corrosion-resistant adhesion layers to the threads and rough electrode materials that increase their effective surface area without increasing their size.
为了优化我们的线与周围组织的相容性，我们认为它们应该与邻近的神经元具有相同的大小规模并且尽可能灵活。螺纹还必须抵抗组织中流体的腐蚀。因此，我们用薄膜金属和聚合物微加工螺纹。为了满足这些标准，我们开发了新的微加工工艺并在材料科学方面取得了进步。其中包括将耐腐蚀粘附层集成到螺纹和粗糙的电极材料中，从而在不增加尺寸的情况下增加其有效表面积。
CHIPS 低功耗集成芯片
Our Link needs to convert the small electrical signals recorded by each electrode into real-time neural information. Since the neural signals in the brain are small (microvolts), the Link must have high-performance signal amplifiers and digitizers. Also, as the number of electrodes increases, these raw signals become too much information to upload with low power devices. Scaling our devices requires on-chip, real-time identification and characterization of neural spikes. Our custom chips on the Link meet these goals, while radically reducing per-channel chip size and power consumption compared to current technology.
我们的 Link 需要将每个电极记录的小电信号转换为实时神经信息。由于大脑中的神经信号很小（微伏），因此 Link 必须具有高性能信号放大器和数字转换器。此外，随着电极数量的增加，这些原始信号变得信息量太大，无法通过低功率设备上传。扩展我们的设备需要对神经尖峰进行片上实时识别和表征。我们在 Link 上的定制芯片满足了这些目标，同时与当前技术相比，从根本上减少了每通道芯片的尺寸和功耗。
HERMETIC PACKAGING 亲肤封装材料研制
The Link needs to be protected from the fluid and salts in the brain. Making a water-proof enclosure can be hard, and it’s even harder when that enclosure must be constructed from biocompatible materials, replace the skull structurally, and allow over 1,000 electrical channels to pass through it. To meet this challenge, we are developing innovative techniques to build and seal each major component of the package. For example, by replacing the connection of multiple components with a process that builds them as a single component, we can decrease device size and eliminate a potential failure point.
需要保护 Link 免受大脑中的液体和盐分的影响。制作防水外壳可能很困难，而当外壳必须由生物相容性材料制成、在结构上替换头骨并允许 1,000 多个电气通道穿过时，就更难了。为了应对这一挑战，我们正在开发创新技术来构建和密封包装的每个主要组件。例如，通过将多个组件的连接替换为将它们构建为单个组件的过程，我们可以减小设备尺寸并消除潜在的故障点。
NEUROSURGERY 高精度神经手术机器人
Our threads are too fine to be manipulated by hand and too flexible to go into the brain on their own (imagine trying to sew a button with thread but no needle). Yet, we need to safely insert them with precision and efficiency. We are innovating on robot design, imaging systems, and software to build a robot that can precisely and efficiently insert many threads through a single 25 mm skull opening while actively avoiding blood vessels on the surface of the brain.
我们的线太细，无法用手操作，也太柔韧，无法自行进入大脑（想象一下，试图用线而不用针缝纽扣）。然而，我们需要精确高效地安全插入它们。我们正在机器人设计、成像系统和软件方面进行创新，以构建一个机器人，该机器人可以通过一个 25 毫米的颅骨开口精确高效地插入许多线，同时主动避开大脑表面的血管。
NEURAL DECODING 精确而自然的神经解码模型
Neural spikes contain a lot of information, but that information has to be decoded in order to use it for controlling a computer. Academic labs have designed computer algorithms to control a virtual computer mouse from the activity of hundreds of neurons. Our device is intended to record from over an order of magnitude more neurons, which we hope will provide more precise and naturalistic control of electronic devices. To accomplish this, we are building on recent advances in statistics and algorithm design to improve the efficacy and robustness of neural decoding. One challenge is to design adaptive algorithms that maintain reliable and robust performance while continuing to improve over time. Ultimately, we want these algorithms to run in real time on the implanted device itself.
神经尖峰包含大量信息，但必须对这些信息进行解码才能将其用于控制计算机。学术实验室已经设计了计算机算法来通过数百个神经元的活动来控制虚拟计算机鼠标。我们的设备旨在记录超过一个数量级的更多神经元，我们希望这将为电子设备提供更精确和自然的控制。为实现这一目标，我们正在建立统计和算法设计的最新进展，以提高神经解码的效率和鲁棒性。一个挑战是设计自适应算法，以保持可靠和稳健的性能，同时随着时间的推移不断改进。最终，我们希望这些算法能够在植入设备本身上实时运行。
