@InProceedings{Peel01,
  title =        "{A} {R}econfigurable {H}ost {I}nterconnection {S}cheme for {O}ccam-{B}ased {F}ield {P}rogrammable {G}ate {A}rrays",
  author=        "Peel,  Roger M. A.",
  editor=        "Chalmers,  Alan G. and Mirmehdi,  Majid and Muller,  Henk",
  pages =        "179--192",
  booktitle=     "{C}ommunicating {P}rocess {A}rchitectures 2001",
  isbn=          "1 58603 202 X",
  year=          "2001",
  month=         "sep",
  abstract=      "This paper reports on the development of an interconnection
     scheme for field-programmable gate arrays (FPGAs). These
     FPGAs may be programmed in the Occam parallel programming
     language. Now, not only may the inter-process communication
     channels provided by Occam be used on-chip, but they may
     also be extended to a host processor using the ubiquitous
     Universal Serial Bus (USB). Bidirectional channels of BYTEs
     are carried along this bus to a host processor (running
     Linux) where they are presented to application code using a
     device driver that provides similar capabilities to the
     standard B004 card link driver. A unidirectional end-to-end
     throughput between Linux processes and FPGA processes,
     across USB, has been measured as high as 1025 kbytes/sec,
     although this rate is only achieved in favourable
     circumstances. Similarly, 410 kbytes/sec may be transferred
     in both directions simultaneously. Unidirectional
     transmission rates of more than 600 kbytes/sec, and
     bidirectional rates of 175-300 kbytes/sec in each direction
     may be achieved in a wide range of circumstances. The paper
     presents a range of performance figures, explaining which
     are limited by the underlying characteristics of the USB bus
     and which are caused by the current implementation. By
     implementing a transputer OS-Link in the FPGA, it is
     possible for a USB- enabled computer to communicate with a
     network of transputers, providing a convenient - and
     potentially faster - alternative to previous methods."
}