[2020] Event-Based Communication Protocol for Ultra-Low Power Compressive Neural Interfaces [Already taken]

Current neural interfaces provide a coarse communication link that does not respect the single-cell specificity of the neural network they are targeting, indiscriminately activating or recording multiple cells at the same time, leading to poor performance when implemented in brain-machine interfaces (BMIs).

The next generation of BMIs will interface to the neural system at single-cell resolution to improve the state-of-the-art. To do so, they will require higher resolution and channel count interfaces. However, such interfaces will generate a massive amount of data. 10,000 channels with 10-bit resolution at 20,000 samples per second generate 2 Gbps of neural data. This would require 125 times the data rate used to stream 4K videos from Netflix - 90% of which contains no information since the spike rate of neurons is much less than the sampling frequency. The energy cost of generating and transferring this large amount of data is prohibitive.

Researchers have investigated a wide range of options to address this challenge, with most solutions proposing to compress the neural data on-chip before transmission. A common issue with compressive neural interfaces is that the data output is event-based and not uniform. This complicates the design of the communication protocol and the transmitter, requiring novel architectures to maintain the benefits of data compression throughout the entire system.

Assignment

The idea is to study an event-based communication protocol for ultra-low power compressive neural interfaces. The protocol will be implemented in CMOS technology and verified post place&route. To verify the power efficiency of the protocol, an existing compressive algorithm and previously recorded neural data will be used.

Requirements

MSc EE-ME student.

You should be comfortable with Verilog/VHDL and the digital flow for IC design (synthesis and place&route using Synopsys/Innovus). Curiosity, hard work, and creativity are always needed. If you are interested, contact Dr. Dante Muratore via email with a motivation letter and attached CV (with taken courses and grades).

Contact

Last modified: 2022-06-23