RELIABILITY AND INFORMATION STORAGE CAPACITY OF SYNAPSES Lav R. Varshney1,2 and Dmitri B. Chklovskii2 1. Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA 2. Cold Spring Harbor Laboratory, Cold Spring Harbor, NY Synaptic plasticity is widely believed to be the main mechanism for long-term memory storage in the brain. We use information theoretic concepts to investigate the properties of synapses that optimize the information storage capacity under volume constraints and find that synapses should be small, and thus unreliable. Furthermore, synaptic weights should be used as binary storage elements in nearly all cases. Although information storage capacity is maximized with small, unreliable synapses, speed of memory retrieval suffers. The competing desiderata for information storage (capacity) and information retrieval (time) cast the neural information storage problem into a classical capacity-time under cost constraints tradeoff as is encountered in electronics. We posit that the wide distribution of synaptic strengths that is observed in the brain suggests that some portions of the brain are fast, robust, but less capacious, and may serve as infrastructural elements, whereas other portions of the brain are slow but more capacious, serving as depositories of information. We also believe that the distribution of synaptic strengths optimizes storage capacity while still meeting the retrieval quality of service requirements that are present for the organism to survive in a complex, fast-paced environment.