ABSTRACT
We describe the optimum design of an execution
stage for embedded processors. The execution stage
is the core of embedded processors to treat basic applications
for multimedia, communication and control.
Considering algorithms in running such applications,
the design principle of embedded processors
is to achieve reliability, precision, low power, high
speed and fast throughput altogether. However, it
is hard to sufficiently clear the design principle by
existing microprocessor technology. The one of difficult
issues is the mixed sequence of integer and FP
(floating point) arithmetic instructions. The different
latencies of these instructions in a processor’s execution
stage surely make the pipelined processing
disturb and degrade throughput. The solution for
this problem in this study is a wave-pipelined MFU
(multifunctional unit) that is the multifunctionalization
and wave-pipelining of different FUs (functional
units). The high speed characteristic of wave-pipeling
guarantees the reliability of processor behavior that is
free from timing error. Thus, the waved MFU agrees
well with overall trade-off design. In order to complement
the shortage of the standard CAD tools, we
explored the HS/SW (hardware/software) co-design
for the waved MFU. Experimental results by using a
0.18-m CMOS standard cell technology showed the
usefulness of this approach.
ABSTRACT
We describe the optimum design of an execution
stage for embedded processors. The execution stage
is the core of embedded processors to treat basic applications
for multimedia, communication and control.
Considering algorithms in running such applications,
the design principle of embedded processors
is to achieve reliability, precision, low power, high
speed and fast throughput altogether. However, it
is hard to sufficiently clear the design principle by
existing microprocessor technology. The one of difficult
issues is the mixed sequence of integer and FP
(floating point) arithmetic instructions. The different
latencies of these instructions in a processor’s execution
stage surely make the pipelined processing
disturb and degrade throughput. The solution for
this problem in this study is a wave-pipelined MFU
(multifunctional unit) that is the multifunctionalization
and wave-pipelining of different FUs (functional
units). The high speed characteristic of wave-pipeling
guarantees the reliability of processor behavior that is
free from timing error. Thus, the waved MFU agrees
well with overall trade-off design. In order to complement
the shortage of the standard CAD tools, we
explored the HS/SW (hardware/software) co-design
for the waved MFU. Experimental results by using a
0.18-m CMOS standard cell technology showed the
usefulness of this approach.
การแปล กรุณารอสักครู่..
