This project lays theoretical and technological foundations for future biohybrid and brain-machine-interface devices, taking advantage of recent progress in our theoretical understanding of brain, in-vivo and in-vitro electrophysiology, and biomimetic hardware components (neuromorphic VLSI). If successful, the project will interface biomimetic hardware devices seamlessly with neural tissues and, what is more, will deliver theories and technologies for future artifacts that integrate, functionally and physically, into functioning brain systems.We hold that the basis of neural representation and processing are discrete and repeating states of collective activity ("attractor states''). Thus, an efficient interface with neural tissue must take into account this spontaneous intrinsic dynamics. We argue that the most efficient and least traumatic way to influence perception, movement, and behavior in general will be to operate by enhancing the probability that collective activity visits certain desired states.Biasing the internal dynamics with gentle perturbations is a key tenet of the proposed project, both as a high road to understanding the dynamical processes unleashed by direct electrical stimulation of neural tissue, and as a philosophical stance on how artifacts should interact with the neural realm. In contrast, the mainstream approach to BMI (brain-machine interfaces) simply inserts electrical signals into the brain, imposing abnormal activity states and disrupting the intrinsic activity dynamics.We will develop the capacity for gently steering the dynamics of neural systems in two complementary experimental set-ups: large-scale networks of cortical neurons cultured in-vitro, and sensory neocortex of awake and behaving rats. By steering the activity dynamics of sensory neocortex it in-vivo, we will also demonstrate the capacity for controlling the animals' performance in sensory choice tasks.A prerequisite for steering is the capacity to recognize dynamical states of neural tissues. A significant theoretical and experimental effort is dedicated to that effect. Once this is achieved, we will be ready to face the ultimate goal of performing the readout by biomimetic networks with intrinsic dynamics over multiple time-scales. When interfaced with neural tissue, such networks can couple with the neural attractor dynamics and state-dependent processing, classifying activity trajectories that are nested over multiple time-scales.To realize biomimetic networks in hardware, the state-of-the-art in neuromorphic VLSI will be advanced through improved "microscropic'' components (integrate-and-fire neurons and synapses), novel "mesoscopic'' components (nodes with multi-stable dynamics), and more realistic connection topologies. Control and communication infrastructures will be subsumed into a largely autonomous system-on-chip ("NeuroSoC'').