Writings on Energy in Computations
Energy is the key limiter in most of our computing systems (battery life
for mobile, power density for all systems). With the end of Dennard
Scaling, we can now place more transistors on a chip than we can afford to
simultaneously use. Consequently, designs that use the least energy
can often provide the highest performance within a limited power density envelope.
Architecture
- Fundamental
Underpinnings of Reconfigurable Computing Architectures (Proc. IEEE,
2015) -- tutorial overview of how programmable architecture (including
FPGAs and processors) impact energy in computations.
[IEEE
Xplore link]
Following papers provide greater details on this topic:
- Location, Location, Location---The Role of Spatial Locality
in Asymptotic Energy Minimization (FPGA 2013) -- asymptotic
analysis of how architecture drives energy consumption.
[Abstract and Paper Link]
- Wordwidth, Instructions, Looping, and Virtualization---The Role of
Sharing in Absolute Energy Minimization (FPGA 2014) -- analytic
models, including constants, on how architecture drives energy
consumption.
[Abstract and Paper Link]
- Energy Minimization in the Time-Space Continuum (FPT
2015) -- why time-multiplexing of bit-level FPGAs doesn't save energy
compared to spatial FPGAs based on concrete design and benchmark mappings.
[Abstract and Paper Link]
- Kung Fu Data Energy---Minimizing Communication Energy in FPGA
Computations -- given that you have a fixed programmable architecture
substrate, how to tune parallelism to minimize energy consumption.
[Abstract and Paper Link]
- Impact of Memory Architecture on FPGA Energy Consumption
(FPGA 2015) -- how to design the memory architecture to minimize the
mismatch energy in programmable architectures.
[Abstract and Paper Link]
Variation
- Limit Study of Energy & Delay Benefits of Component-Specific
Routing (FPGA 2012) -- how much energy benefit can we get using
post-fabrication mapping to tolerate high variation in the
programmable interconnect of FPGAs?
[Abstract and Paper Link]
- More details in ``Messy'' Computing theme.
Transients
- Energy Reduction through Differential Reliability and Lightweight
Checking (FCCM 2014) -- how to run most of the computation at
low energy (low voltages) and detect errors reliably and inexpensively.
[Abstract and Paper Link]
- Final Report of the CRA/CCC Visioning study on Cross Layer
Reliability (2011) -- vision and research agenda for addressing
reliability and energy in highly-scaled technologies. [links to report]
André DeHon