16
ASET

ALGORITHM LEVEL ANALYSIS and DATA CORRELATION

Embed Size (px)

Citation preview

ASET

ASET

The sources of power consumption on a CMOS chip can be classified as

dynamic power, short circuit and static current. But at algorithm level we are

consider only dynamic power and rest are determined at the circuit level.

Using the smart circuit techniques ,we can reduce the dissipation due to short

circuit currents to less than 15% of the entire power.

The power dissipated can given by the equation,

Power = Ceff V2f

Ceff = aC

Algorithm level analysis are to be used for guiding design as its help in

magnitude improvement, accuracy and give accurate relative information.

ASET

POWER DISSIPATION

Algorithm Inherent

Dissipation

Implementation

Overhead

The AID comprises the power of the execution units and

memory. This is inherent in the sense that it necessary to achieve

the basic functionality of the algorithm.

The IO includes control interconnect and registers. The power

consumed by this component depends largely on the choice of

architecture and implementation.

ASET

The AID refers to the power consumed by the execution units and memory. This

component is fundamental to a given algorithm and is the prime factor for comparison

between different algorithms as well as for the qualifying the effect of algorithm level

design decisions.

Its dissipation can be estimated by a weighted sum of the number of operations in

the algorithm. The weights used for the different operations must reflect the respective

capacitances switched.

To estimate the capacitance switched at this level it is necessary to ignore the

statistics of the applies data and to employ a white noise model i.e. random data.

Switching statistics are strongly influenced by the hardware architecture. The mapping

of operators onto the hardware resources effects the temporal correlations between

signals.

ASET

The implementation overhead consists of the control Interconnect and

implementation related memory/register power. The power consumed by these

components depends on the specific architecture platform chosen and on the

mapping of the algorithm onto the hardware.

Since this overhead is not essential to the basic functionality of a given algorithm,

several estimation tools ignore its effect for algorithm level comparisons. The

power consumed by these components is often comparable if not greater than the

AID.

So, its important to get reasonable estimates of the implementation overhead for

realistic comparisons algorithms and to guide high level decisions.

ASET

The implementation dependant power consumption is strongly correlated to a number of

properties of the algorithm. For example, structural property of an algorithm may affect

the amount of his overhead is the locality of reference.

SPATIAL LOCALITY TEMPORAL LOCALITY

1. It refers to the extent to which an

alog. can be partitioned into natural

clusters based on connectivity.

2. A spatially local alog. renders itself more

easily to efficient partitioning on

hardware, allowing highly capacitive

global buses to be used sparingly.

3. In terms of memory/registers access,

spatial locality refers to the distance

between the address of items referenced

close together in time.

4. A spatially local memory access pattern

allows portioning of memory into smaller

blocks that require less power per access.

1. It refers to the average lifetimes of

variables.

2. It tends to require less temporary

Storage and have small registers files

leading to lower capacitances.

3. It refers to the probability of future

access to items referenced in the

past.

4. Its does not allow any portioning

of memory into smaller blocks.

ASET

After examining methods for estimating power consumption at the algorithm level,

The next logical step is to examine power minimization techniques at this level.

The general approaches for power minimization are:

1. Voltage Reduction: At the algorithm level, functional pipelining, retiming, algebraic

transformations and loop transformations can be used to increase the speed and

allow lower voltages but these approaches often translate into larger silicon area

Implementations and hence this approach is called as trading area for power.

2. Avoiding Wasteful Activity: At the algorithm level the size and complexity of a given

algorithm(operation counts word lengths) determine the activity. If there are several

algorithm for a given task the one with least number of operations is preferable.

ASET

This category includes:

Operation Reduction : It includes common sub expression elimination, algebraic

transformations, dead code elimination.

Strength Reduction: It refers to replacing energy consuming operations by a

combination of simpler operations. The most common in this category is expansion of

multiplications by constraint into shift and add operations. Its results in low power but

sometimes it results opposite if there is increase in the critical path.

Memory Path: Algorithm level transformations for minimizing the power

consumption of memories present in. These include conversion of background

memory to foreground register files and reduction of memory size using loop

reordering and loop merging transformations.

ASET

It is a more challenging problem as these algorithm posses structural properties

such as locality and regularity and we have to detect these properties.

Spatial Locality can be detected and used to guide portioning . Regular algorithm

typically require less control and interconnect overhead.

Other way to reduce the implementation overhead is to reduce the chip area as

this typically translates into reduced bus capacitances.

ASET

ASET

ASET

ASET

The effect of data correlation on power dissipation often depends on the

numerical representation of the digital system.

If the data sample is positively correlated, successive data sample values

are very close in their binary representation.

UNIFORM WHITE NOISE REGION: If the LSB bits toggle at

approximately half the maximum frequency as bits are toggle in the random

fashion.

on MSB side, the bits have a very low toggle rate and they are called the

SIGN BIT REGION as most of the toggling in this region are sign change

and have very high Switching Frequency.

GREY REGION is the area between two regions where the toggle

frequency changes from WHITE NOISE to SIGN BIT.

ASET

ASET

ASET