C h a p t e r 3
CPUs, GPUs, and FPGAs span a wide
spectrum that requires different voltage
rails depending on purpose. Pinpointing
certain power management requirements
individually is difficult with so many
different devices.
Most FPGA devices are found in higher-
end computing applications and require
many different supply voltages. In data
centers, FPGA-based accelerators
are growing in demand to support
workloads such as financial modeling,
machine learning, genomics, and video
transcoding. Beyond the data center,
FPGAs are a key component of 5G
infrastructure. Both environments are
increasingly densely packed, putting
further pressure on both power and
thermal management, while also having to
be highly responsive.
Even as CPUs have become increasingly
powerful, GPUs have become the
processor of choice for AI workloads—
their heritage in graphics rendering
makes them ideal for performing highly
parallel computations and processing
large amounts of data simultaneously.
FPGA diversity adds power
management complexity
The shrinking size of processors and
FPGAs has made power management
more challenging, especially with
increasingly lower voltage and a high-
POWER PROCESSORS
AND FPGAs
Modern FPGA tends to operate at a
lower supply to reduce dissipation,
which makes the supply design
job harder. The most important
consideration is to design an
accurate and stable voltage supply
for the FPGA."
Hao Jie Chan
Senior Analog Hardware Engineer,
NI (now Emerson)
17
Power Management: Efficiently Powering Processors, FPGAs, and Microcontrollers
