The chips are created by Intel, one of the project's commercial partners. The team is also using Intel's open-source compilation library, nGraph, which allows neural network models to be compiled to different hardware targets. This means that the same network code can be run on a server with a lot of power or compiled down to run on something with much less processing power, like a phone or, perhaps one day, an implanted device.
"nGraph gives us a unique pathway to pilot possible future therapeutic models across a wide range of hardware targets. The hope is we can compile our algorithms down to an implantable system in the future, without significant redesign," says Borton.
In the upcoming project, the initial research will focus on exploring what signals remain to be recorded in the spinal cord post-injury, and how those signals could be used to control the legs for walking, standing, and signals related to bladder control, which is a primary concern reported by people with spinal cord injuries. The project aims to demonstrate that the spine-to-spine device can properly target the neural circuits that influence these activities.
During the initial phase, an external computer will connect the upper and lower spine and decode the signals, similar to the computer-in-the-loop systems utilized in Borton's earlier neural device work. But the researchers hope this study will lay the groundwork for a future fully implantable device. The first human studies are expected to start in spring 2020.
"The project is a big step from a scientific perspective. We simply don't know everything about what the spinal cord does, particularly post-injury," says Borton. He compares the limited knowledge of the spinal cord to the state of our knowledge of the brain 20 years ago. Thanks to increased funding and interest in brain research, that knowledge has grown substantially in recent years.
Intel - www.intel.com
Brown University - www.brown.edu