You can download my thesis from this link. This work was done under supervision of Prof. Euisik Yoon at the University of Michigan. In other posts in this website, I will try to summarize some of the topics covered in this work. Below you can read the abstract.
Understanding dynamics of the brain has tremendously improved due to the progress in neural recording techniques over the past five decades. The number of simultaneously recorded channels has actually doubled every 7 years, which implies that a recording system with a few thousand channels should be available in the next two decades. Nonetheless, a leap in the number of simultaneous channels has remained an unmet need due to many limitations, especially in the front-end recording integrated circuits (IC).
This research has focused on increasing the number of simultaneously recorded channels and providing modular design approaches to improve the integration and expansion of 3-D recording microsystems. Three analog front-ends (AFE) have been developed using extremely low-power and small-area circuit techniques on both the circuit and system levels. The three prototypes have investigated some critical circuit challenges in power, area, interface, and modularity.
The first AFE (16-channels) has optimized energy efficiency using techniques such as moderate inversion, minimized asynchronous interface for data acquisition, power-scalable sampling operation, and a wide configuration range of gain and bandwidth. Circuits in this part were designed in a 0.25μm CMOS process using a 0.9-V single supply and feature a power consumption of 4μW/channel and an energy-area efficiency of 7.51×1015 in units of J-1Vrms-1mm-2.
The second AFE (128-channels) provides the next level of scaling using dc-coupled analog compression techniques to reject the electrode offset and reduce the implementation area further. Signal processing techniques were also explored to transfer some computational power outside the brain. Circuits in this part were designed in a 180nm CMOS process using a 0.5-V single supply and feature a power consumption of 2.5μW/channel, and energy-area efficiency of 30.2×1015 J-1 Vrms-1mm-2.
The last AFE (128-channels) shows another leap in neural recording using monolithic integration of recording circuits on the shanks of neural probes. Monolithic integration may be the most effective approach to allow simultaneous recording of more than 1,024 channels. The probe and circuits in this part were designed in a 150 nm SOI CMOS process using a 0.5-V single supply and feature a power consumption of only 1.4μW/channel and energy-area efficiency of 36.4×1015 J-1Vrms-1mm-2, which is the highest reported efficiency to date.