Tamarin
Goals:
Create a base of Lisp machine subsystems to evolve a sequence of machines
Build an initial VLSI based Lisp Machine.
Totally custom VLSI
Low cost
Fast completion
Desk Top form factor
Medium performance
Research issues:
Process of quickly building systems
NOT technological issues
Methodology:
Evolvable
Basis for future machines
Avoid too much generality - make it specific to Lisp
Stay close to product issues
Simplify at every oppotunity
Where possible, compromise and finesse instead of directly solving technological issues
Eliminate unknowns early - ??????
Use system perspective
Use existing silicon pieces where possible
Design it to make Interlisp software simpler and get around Lisp’s bottlenecks
Performance leverage:
Lisp has mismatches with current architectures
16 bit vs 40 bit machines
typed vs. untyped words
IFU mismatch
Lack of usable hardware stack
Close memory coupling
Lack of hardware virtual memory
Architectural details
3 custom chips
Memory interface
Lisp processor
DMA/IO controller
Ethernet
SCSI interface for disk
Display
IBM PC bus
1 PC board
Current Status
Memory Interface chip completed - waiting for fab
PC board design completed, fast turn fab building it
Floorplan of Lisp chip - first pass completed
Architecture of IO processor underway
Comparason with Dragon project
Trying to create evovable set of machines
Sooner timescales
No cache
Avoiding multiprocessor research issues
Lower on complexity curve
Physically smaller machine
Much smaller effort