Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Yaseer A. Durrani
Teresa Riesgo
I. INTRODUCTION
During the years of inception, the use of integrated circuits
(IC) was confined to traditional digital electronic systems
such as wearable computers, wireless communication
systems. Nowadays not only those devices play an
increasingly important role but also the use of integrated
systems is much more widespread, from controllers used in
home appliances to the automobile industry. The digital
electronic circuits are becoming more application specific.
The shrinking of devices due to the development of new
fabrication technology has increased dramatically the
number of transistors available for use in a single chip. The
larger capacity of the chips is also being used to extend the
functionality of the systems.
However, the importance of low power dissipative digital
circuits is being increased rapidly. In order to handle the ever
increasingly complexity, CAD tools have been developed.
Those tools also help minimizing power dissipation of digital
devices and accurate power estimation tools are needed at
high abstraction levels.
In response to this need, power macromodeling technique
is a promising solution to the problem of high-level power
estimation. The macromodel is to generate a mapping
between the power dissipation of a circuit and certain
statistics of input signals. This technique has been proven to
be effective for individual IP components [1]. The urgent
need of a feasible IP power model is becoming more useful
in recent years. The application of power macromodeling on
IP blocks of the system requires knowledge of the signal
statistics among different IP blocks. To obtain this
information, the architect must perform different functional
simulations.
In this paper, we focus on the problem of power estimation
at register transfer level (RTL) for IP-based designs. Various
power estimation techniques have been introduced
previously. These techniques can be divided into two
1-4244-0136-4/06/$20.00 '2006 IEEE
4935
(1)
Pin
i=1
j =1 ij
s 1
j =1
i =1
Input Pattern
qij qi +1 j
r (s 1)
r
j =1
k =1
i =1
(3)
q ij qik
s r (r 1)
(2)
Tin =
(9)
rs
Din =
S in
s t +1
j =1
t 1
RTL Circuit
(4)
(yj qj )
rs
Power
Libraries
(5)
Power Simulator
Operation for
generating new
patterns
No
Convergence ?
Yes
Average Power
Estimation
(6)
(7)
(8)
4936
Circuits
f A , f B , f C and f D that maps the input metrics of a macroblock to its output metrics Pout, Dout, Sout, Tout. The input
sequences we use are highly correlated generated by our new
method. The power is estimated using Monte Carlo zero
delay simulation technique. The power values predicted by
LUT are compared with those from simulations, and average
error and maximum errors are computed.
Experimental results show that our generated sequences
are with accurate statistics and high convergence. For the
input metrics, Pin, Din, Sin, Tin we specify the range between
[0.1, 0.9]. We generated 550 sequences with 8, 16, 32 bits
wide. The sequence length is 2000 and 1000 vectors for
macro-blocks.
Average Error
Max Error
Mult8x8-1
0.76%
2.36%
Mult8x8-2
1.53%
3.02%
Mult8x8-3
0.60%
2.99%
Mult4x4-1
0.65%
2.31%
Mult4x4-2
2.15%
3.00%
Mult4x4-3
2.92%
4.70%
Mult4x4-4
0.94%
2.44%
Mult4x4-5
10.07%
11.31%
Mult4x4-6
1.21%
2.83%
Comp-1
1.60%
2.96%
Comp-2
0.29%
1.10%
Comp-3
0.73%
1.57%
Comp-4
0.47%
0.66%
10.6
4937
10.4
A.
10.2
10
9.8
9.6
9.4
9.2
9
9.5
10
Power from Macromodel
10.5
B.
1.35
Steady State
P o w e r (m W )
1.3
1.25
1.2
1.15
1.1
1.05
100
200
300
400
500
600
700
Interval Length (Clock Cycle)
800
900
V. CONCLUCIONS
1000
10
9.5
VI. REFERENCES
12
Steady State
11.5
11
P o w e r (m W )
Convergence Analysis
10.5
[1]
8.5
200
400
600
1600
1800
2000
[2]
52
Steady State
[3]
50
P o w e r (m W )
48
46
[4]
44
42
40
[5]
38
36
200
400
600
1600
1800
2000
[6]
4938
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
4939