You are on page 1of 5

Session-11

Architecture level analysis


• Over the years, the design abstraction has moved from the mask, the transistors, to
gates and now to the architecture level
• Architecture level abstraction is called as block level or macro level design
• Building blocks
– Registers
– Adders
– Multipliers
– Buses
– Multiplexers
– Memories
– State machines etc
• Today, architecture level power analysis is becoming more important because more
digital circuits are now synthesized from architecture description
Power model based on activities
• One way to characterize the architectural components is
to express the power dissipation as a function of the
number of bits of the components and their operating
frequencies
• Eg: the power dissipation of an adder can be expressed as:
P = (n K1 + K2) f
Where n – number of bits
f – frequency of the addition operation
K1 and K2 – empirical coefficients derived from gate level
simulation
• The model depends only on the operating
above
frequency and size of the adder
• The model does not account the data dependency of the
power dissipation
Power model based on activities
• A more accurate model that can capture data dependency
is to characterize the power dissipation as

P  Ki fi
input i

• For multipliers, the model is given by

P  Ki fi 

K j f j input i input j
 f
P  K1  fi  K2 input j
• Since it is tedious toinput
characterize K j for each input i, the P
i
P  K1 fin i
K2 fout
can be simplified as follows
Power dissipation based on Component operations
• In this model the power dissipation is expressed in terms
of the frequency of some primitive operations of an
architecture component
• The power dissipation is given by

P  K1 fread  K 2 fwrite
• The parameters fread are the frequencies
and fwrite
READ
of and WRITE operations, respectively
• Coefficients K1 and K2 are obtained from characterization
and properties of the component
Abstract statistical power models

• The power models have to make


previous
assumptions on the
certain
statistical behavior of components in
order to model the power dissipation efficiently
• There are several fundamental issues in
composing a power model
– Whichparameters have major influences on the
power dissipation ? Size, signal frequency, operation
frequency etc
– What is the relationship of the power dissipation with the chosen
parameters? Equation or lookup table
• The solution depends on the type, implementation
and accuracy level tolerated by the analysis

You might also like