Our objective is to develop a billion-site cellular automata machine optimized for lattice-gas simulation of geophysical processes. This machine will be a medium-scale implementation of the margolus cam-8 architecture, developed at the mit laboratory for computer science with substantial DARPA funding. A number of smaller CAM-8 are already in use at MIT and collaborating institutions. CAM-8 is a very fine-grained, parallel, uniform, scalable multiprocessor architecture explicitly aimed at the fine-grained modeling of spatially-extended systems. It is meant to complement, rather than displace, more traditional modeling approaches, and is ideally suited for lattice-gas models of physical processes. The current modular design can be directly used for a CAM-8 unit of a billion sites. Our development effort will focus on size-related implementation issues (reliable interconnection of a large number of modules, adequate cooling, high-speed I/O); manufacturing, assembly, and testing; systems and applications software; documentation. Phase I will deliver complete building plans (VLSI chips, schematics, board artwork, hardware manuals) for a giga-site unit, basic systems software, and an outline of additional software. It will also provide a demonstration of multi-module interconnection (256 mega-sites) and of visualization techniques. Phase II will carry out the plans, with modifications and improvements subject to negotiation.