Fully Integrated Systems for Neural Signal Recording: Technology Perspective and Low-Noise Front-End Design

Since the dawn of microelectronic industry integrated technologies have been fuelling tremendous advances in science, engineering and applications leading to an increasing inclusion of intelligence in infrastructures, equipments and products. This trend, leveraging on silicon device miniaturization, is still on-going and is having a profound impact in all fields, medical science and therapeutics included. In the forthcoming years, availability of decananometer silicon technologies, advances in micro-mechanical and packaging manufacturing, energy-conversion techniques and material engineering are expected to provide the solutions needed to develop fully miniaturized, low-power, energy- autonomous smart systems. These systems will promote a more intimate smart link between humans, from a high level interaction down to cellular level, $"$things$"$ and environment. Implantable recording systems are a challenging test field for deeply scaled technologies since demanding performance required for the application and the tight constraints imposed by the surrounding environment, i.e. the body. But big challenges translate in big opportunities: the potentials of this trend are already clearly visible, neurotechnology being one of the leading examples. Technological advances are enabling innovative interfaces between neurons and electronics, opening the way to new therapeutic devices for neurological diseases as well as to detailed investigation tools of the cognitive processes. The chapter reviewed the performance requirements and the perspectives of fully integrated neural recording systems, pointing out the issues faced in the definition of optimal architectures and function partitioning. In this frame, energy-efficiency and low noise design are key ingredients. The fundamental metrics to quantitatively judge the trade off between noise, power consumption, and processing speed have been introduced and adopted to compare the most recent system implementations. It has been shown that a 10uW power budget target per sensing channel is attainable by using cutting edge technologies and careful design. Finally, we focused on the particular issue of neural amplifier design: leveraging on a detailed break-down of the noise sources and by mean of an insightful design strategy we addressed the problem of noise-power trade-off and we presented a neural amplifier that achieved the best performance so far reported.