The goal of the Cortical Testbed is to restore higher cognitive functions that are lost as a result of damage (stroke, head trauma, epilepsy) or degenerative disease (dementia, Alzheimer’s disease). Specifically, we are focusing on loss of long-term memory formation which is supported by the hippocampus and surrounding limbic cortical brain regions. Unlike the first 2 testbeds, this testbed does not have a prototype device in clinical trials and is thus further from a clinical application. However, our group has made significant progress and this testbed has the added benefit of contributing to the overall effort of neuralprosthesis. Achieving a neural prosthesis in this case requires replacing damaged hippocampus with biomimetic devices that (1) perform the same signal processing functions (biologically realistic, nonlinear input/output transformations) of the damaged neurons, and that (2) bi-directionally communicate with the afferents and efferents of the damaged hippocampal circuits to functionally “by-pass” the lost/dysfunctional brain area(s). During the past three years, we have successfully demonstrated a “proof-of-principle” of this approach using a hippocampal slice preparation. Specifically, we have shown that we can replace one portion of the multi-component circuit of the in vitro hippocampus with a VLSI-based model that accurately predicts the nonlinear dynamics of that component, and as a consequence, restores hippocampal system function. Completing this “proof-of-principle” allowed us to develop the fundamental strategy and tools to achieve a hippocampal prosthesis at a single circuit level. Already during the past year 3 and this year 4, we are building on these accomplishments to develop a multi-circuit, systems level solution for a prosthesis. This device is designed for the behaving rat, i.e., designing and fabricating novel penetrating electrode arrays that conform to the cytoarchitecture of the three-dimensional hippocampus, recording the propagation of electrophysiological activity of hippocampal neurons during learning and memory functions of the behaving animal, developing a multi-input/multi-output model that captures the essential nonlinear dynamics of the hippocampal system, implementing that model in VLSI, and integrating the components of this neural prosthetic system to restore memory function in an animal with hippocampal damage. In the process of development, we have also identified a more near-term application for the cortical testbed. We plan to explore extending the approach described above to the control of hippocampus-generated epileptiform activity, i.e., pharmacologically or electrically suppressing an epileptic focus, and then by-passing that suppressed brain region with a biomimetic model to maintain cognitive function.