The Defense Department has hired one of the nation's top computer scientists from Xerox Corp. to help lead the federal research effort to beat Japan in the development of a new supercomputer.

Lynn Conway, manager of the Knowledge Systems Area at Xerox's Palo Alto Research Center, will become computer research manager at the Defense Advanced Research Projects Agency, which has been given responsibility for the supercomputer development program. The agency has asked for $50 million for its 1984 supercomputer budget and an additional $95 million for 1985. The supercomputer program will be administered through DARPA's Information Processing Techniques Office.

The supercomputer concept, which has been dubbed "fifth generation," is technologically a great leap from the current most sophisticated computers. First came vacuum tubes, then transistors, then computer chips. The fourth generation of computers uses chips with very large scale integrated circuits, referred to as VLSI. The fifth-generation supercomputer will use VLSI as its hardware foundation.

Conway, an electrical engineer, is considered one of the pioneers of VLSI design. Her group at Xerox also has been exploring ways to apply artificial intelligence techniques to VLSI design. Several computer systems experts consider such efforts critical, because designing chips that can handle millions of logical functions is increasingly complex for human designers without automated help.

The DARPA fifth generation effort is considered by many--in and out of the Defense Department--as a direct response to Japan's own supercomputer project. "I don't want to say it's a fifth- or sixth-generation machine," Richard D. DeLauer, Defense undersecretary for research and engineering has said, "only that it will outperform anything the Japanese develop."

The Japanese effort, sponsored by its Ministry of International Trade and Industry and tentatively funded at half a billion dollars, is designed to establish Japan as the world leader in computer technology by the end of this decade.

DARPA's supercomputing program, which is called "Strategic Computing and Survivability," started earlier this year. The thrust of the program--the emphasis on hardware versus software, for example--is still undetermined, several DARPA sources say. Conway, whose duties begin in August, is expected to help define priorities in the DARPA program. According to a senior staff member of the House Armed Services Committee, funding for the program is contingent upon a development agenda.

There are two aspects of supercomputing that will be addressed by the DARPA program. The first is the effort to make computers calculate thousands of times faster than they can today. Such "number-crunchers" could be used for such numerically dense applications such as long-range weather forecasting.

The other, more ambitious aspect is to build computers that meld advanced hardware with sophisticated software techniques to create machines that calculate better as well as faster. Such machines might apply "expert systems" software and reason to solve complex problems much the same way human experts do. This "expert systems" area is one Conway has been researching at Xerox.

Conway declined yesterday to comment on her DARPA position.