Alain J. Martin
5 references under this name, matched as it is written. Somebody else may write under it too, and the same person may appear here spelled another way.
Alain J. Martin
Alain J. Martin, Caltech. Five records, acquired in one sequence from the institute's repository, and together they run a complete research programme from method to limit to silicon and back to theory. The corpus holds few authors this way.
1985 martin1985delayinsensitive a fair arbiter, derived and proved fair 1986 martin1986compiling the method: CSP-like programs compiled into delay-insensitive circuits 1990 martin1990limitations the limit: entirely delay-insensitive circuits are a nearly empty class 1993 martin1993tomorrows the demonstration: the first asynchronous microprocessor, working on first silicon 1995 manohar1995quasidelayinsensitive with Manohar: the compromise recovers Turing-completeness, and forbids an arbiter
**Read in order, the striking thing is that he published the boundary of his own programme.** martin1986compiling argues that a semantics-preserving translation makes the resulting circuit correct by construction. Four years later martin1990limitations proves that the class of circuits the method targets -- entirely delay-insensitive ones -- admits only C-elements as multiple-input operators and is therefore nearly useless, and names the **isochronic fork** as the weakest assumption that makes design possible again. Most authors leave that to critics.
**And the compromise is then priced, twice.** manohar1995quasidelayinsensitive shows the isochronic fork buys back everything computable, by constructing a QDI Turing machine -- and costs the arbiter, since QDI computations are provably deterministic. martin1993tomorrows shows what it costs in an actual chip: "with the exception of isochronic forks and the interfaces with the memories, the chips are entirely delay-insensitive". An impossibility, a named assumption, a theorem about what the assumption recovers, and the exact boundary in fabricated silicon. The corpus cannot follow any of its other threads that far.
**The 1985 arbiter is the subtle one** and should not be read as contradicting the 1995 result: it builds on a mutual-exclusion element taken as a primitive, which the paper itself describes as making a non-deterministic choice. It is also the strongest thing in derived-not-checked, because the property carried across the transformations is fairness -- a liveness property -- and it settled a question that had been conjectured the other way.
**His counterpart in this corpus is ebergen1989translating**, and the two lines are connected rather than parallel: Ebergen names Martin's technique as similar to his own, and Rem sent Ebergen to Caltech in 1982 for Chuck Seitz's course on self-timed systems. Where Martin compiles by transformation, Eindhoven translates a program syntax into fixed basic elements on the trace-structure formalism of snepscheut1983trace and udding1984classification.
The Caltech deposits hold twenty-four of his reports, 1982 to 2003, including an asynchronous microprocessor in gallium arsenide, low-energy asynchronous memory, and performance analysis of asynchronous circuits. Nothing beyond the five above is held here.