glitch reduction
TRANSCRIPT
-
7/30/2019 Glitch Reduction
1/6
33rd Design Automation Conference Permission to make digital/hard copy of all or part of this work forpersonal or class-room use is granted without fee provided that copiesare not madeor distributed for profit or commercial advantage, thecopyright notice, the title of the publication and its date appear,and notice is given that copying isby permission of ACM, Inc. Tocopy otherwise, to republish, to post on servers or to redistribute tolists, requires prior specific permssion and/or a fee.DAC 96 - 06/96 Las Vegas, NV, USA 1996 ACM, Inc. 0-89791-833-9/96/0006..$3.50
Glitch Analysis and Reduction in Register Transfer Level Power Optimization
Anand Raghunathan
Department of EEPrinceton UniversityPrinceton, NJ 08544
Sujit Dey
C&C Research LabsNEC USA, Inc.Princeton, NJ 08540
Niraj K. Jha y
Department of EEPrinceton UniversityPrinceton, NJ 08544
ABSTRACT: We presentdesign-for-low-powertechniques based
on glitch reduction for register-transferlevel circuits. We analyze
the generation and propagation of glitches in both the control
and data path parts of the circuit. Based on the analysis, we
develop techniques that attempt to reduce glitching power con-
sumption by minimizing generation and propagation of glitches
in the RTL circuit. Our techniques include restructuring multi-
plexer networks (to enhance data correlations, eliminate glitchy
control signals, and reduce glitches on data signals), clocking
control signals, and inserting selective rising/falling delays. Our
techniques are suited to control-flow intensive designs, where
glitches generated at control signals have a significant impact on
the circuits power consumption, and multiplexers and registers
often account for a major portion of the total power. Application
of the proposed techniques to several examples shows significant
power savings, with negligible area and delay overheads.
I. Introduction
Most savings in power consumption can be obtained through acombination of various techniques at different levels of the designhierarchy. We focus on techniques to reduce average power con-sumption in register-transfer level (RTL) circuits. Power estimationtechniques for RTL designs, and high-level synthesis techniques forreducing powerconsumptionhavebeenpreviouslyinvestigated [1, 2].
Several studies have reported the importance of considering glitchingpower during power estimation and optimization [3, 4]. However,very few automated design and synthesis techniques exist for reduc-ing glitching power consumption. At the architecture and behaviorlevels, previous work on power estimation and optimization ignoresthe effects of glitch generation and propagation across the bound-aries of blocks in the architecture. While accurate library modelingapproaches can be used to account for the effect of glitches withinarchitectural blocks, they typically assume that inputs to these blocksare glitch-free. Most previous work at the architecture and behaviorlevels has also sought to focus on data-flow intensive designs, wherearithmetic units like adders and multipliers account for most of thetotal power consumption. However, our experiments with control-flow intensive designs reveal that functional units may constitute amuch smaller fraction of total power than multiplexer networks andregisters.
In this paper,we analyzethe generationand propagationof glitchesin both the control and data path parts of the circuit, and proposetechniques to reduce glitch power consumption. In order to minimizethe generation and propagation of glitches from control as well asdata signals, we propose several techniques including restructuringmultiplexer networks, clocking control signals, and insertingselectiverising/falling delays. These techniques do not rely upon the existence
Supported by NEC C&C Research Labsy Supported by NSF under Grant No. MIP-9319269.
-
OUTPUT
X Y
XIN XY YIN
ZERO
C20
C20
contr[0] = x1 + x3
contr[1] = x0
contr[2] = x0 + x1.c11 + x3.c10
contr[3] = x0 + x1.c11
contr[4] = x0 + x1.c11 + x3.c10
contr[5] = x1.c11.c15 + x2.c15 + x3.c10.c15
contr[6] = x0 + x4
contr[7] = x1.c11 + x2 + x3.c10
contr[8] = x1.c11.c15 + x2.c15 + x3.c10.c15
contr[9] = x0 + x1.c11.c15 + x2.c15 + x3.c10.c15 + x4
c11 = c9 + c10
-
7/30/2019 Glitch Reduction
2/6
-
7/30/2019 Glitch Reduction
3/6
decode logic (see Figure 1) are fed by the outputs of comparatorsand the state flip-flops of the controller. The previous subsectionhas already demonstrated that outputs of comparators can be glitchy.The glitches at comparator outputs can propagate through the decodelogic andcause glitches on thecontrol signals. In addition, the decodelogic can itself generate a lot of glitches, as shown next. Let us focuson control signal contr[2] in the GCD RTL circuit, which is highly
glitchy according to thestatisticsofTable 2. Theportion of thedecodelogic that implements this control signal is shown in Figure 4(a).We observe that though the inputs are nearly glitch-free, significantglitches are generated at AND gates G 1 and G 2. After careful analysis,
x 0
x 1
c 1 1
x 3
c 1 0
contr[2]
22/22
22.5/22.5
/5049.5
/50.5 49.5
49.5/19.5
22.5/2.5
72/20G1
G2
54/53.5
G1G3
(a) (b)
Figure 4: (a) Implementation of control signal contr[2], and (b)
generation of glitches at gate G 1
the generation of glitches atG
1 was attributed to two conditions that
are depicted graphically in Figure 4(b):
C1: A rising transition on signal x 1 was frequently accompaniedbya falling transition on c 11. Thus, the rising transition on x 1 andthe falling transition on c 11 are highly correlated.
C2: Transitions on signal x 1 arrive earlier than transitions on c 11.
Condition C1 arises due to the functionality of the design: most of thetimes when state s 1 is entered (rising transition on x 1), comparatorsfeeding c 9 and c 10 produce a 0, changing from 1 in the previous state.On the other hand, condition C2 is a result of the delay/temporalcharacteristics of the design. A similar explanation holds for theoutput of gate
G
2 being glitchy. Generation of glitches in the controllogic has been described in detail in [7].
IV. Glitch Reduction Techniques
In this section, we describe our techniques for reducing glitchpower consumption in RTL circuits, by minimizing the generationand propagation of glitches through different blocks of the circuit.
A. Reducing glitch propagation from control signals
As shown before, control signals to the data path can be veryglitchy. Ouraim is tostop glitcheson controlsignalsfrom propagatingas close to their source as possible in order to reap the maximum ben-efits in terms of power savings. We illustrate each of our techniquesseparately through examples in this subsection, and later integratethese techniques into a single power optimization framework.
Y