OCV AND TYPES OF OCV ANALYSIS , AOCV,POCV,SOCV
OCV AND TYPES OF OCV ANALYSIS , AOCV,POCV,SOCV
1.1.1
LOGIC SYNTHESIS
The translation of hardware descriptive
RTL code to technology dependent logic gates is called logic synthesis. Gate
level representation of design is termed as netlist. Logic synthesis is done
using Synopsys Design Compiler tool. It includes the following main tasks:
reading in the design, setting constraints, optimizing the design, analysing
the results Slew/Load Dependency Accuracy Improvement constraints defined in
SDC file. DC achieves this by using various optimization techniques enabled by
us for meeting those design constraints.
1.1.2 ON-CHIP
VARIATIONS (OCV)
Process:
It cannot be assumed that all transistors in a chip
would have same process. There may be variations in oxide thickness, channel length, metal
thickness doping concentration etc. due to imperfections during manufacturing
process like mask print, etching etc.
Voltage:
The supply
voltage reaching to the power pins, slew/load accuracy improvement shall not be
the same across all the standard cells. There exists a finite resistance for
the power network. Consider two cells, one which is placed farther, and the
other placed nearby.
Due to
this, the interconnect length for the cell which is placed farther is more, it
will have more resistance and will result in more IR drop, thereby reducing the
supply voltage which is reaching to the cell. Now, the cell placed closer will
have lesser delay as compared to cell placed farther which has lesser voltage.
Temperature:
The
density of transistors is not uniform within a chip. Some of the sections of
the chip may have higher density and higher switching which results in a higher
amount of power dissipation. Therefore local hot spots are formed as the
junction temperature at these sections
of the chip are higher. This result in different delays due to the variation in
junction temperature across the different sections of the chip.
Accounting for
above differences of PVT, there is a possibility that the two regions of the chip
are not at same or exact identical PVT conditions.
There may be many
factors due to which these differences may arise, [3] including:
·
Local
power supply getting affected due to the variation in IR drop along the area of
die.
·
The
NMOS and the PMOS devices having variation in their threshold voltage.
·
The
PMOS or the NMOS device having channel length variation.
·
Existence
of local hot spots resulting in temperature variations.
·
The
interconnect capacitance or resistance getting impacted due to variations in thickness
or interconnect metal etch.
The above together
are referred to as On-Chip Variations (OCV) and can
affect the cell delays and the wire delays in different parts of the chip.
The net delays or
cell delays both can be derated (by making the paths faster or slower) to model
the effect of OCV and hence validate the behavior of the design with these
variations.
1.1.3 EXAMINING
OCV DERATING FOR A SETUP CHECK
Figure 1.1: For Understanding Derating For Setup Check
Typically
the effect of OCV is more pronounced on the clock paths as they travel longer
distances in a chip. Incorporating the OCV analysis while performing STA at any
stage of VLSI flow, is the one way to account for the local PVT variations [3].
PVT
conditions to be slightly different for the capture and launch paths. The way
to include the effect of OCV while performing STA is to derate the delays of
some specific paths, that is by making those paths slower or faster and hence
hereby validating the design behavior along with these variations.[3]
For
setup check, the worst condition occurs when the data clock path and the launch
path have the OCV conditions such that resulting in the largest delays, while
the capture clock path has the OCV conditions such that resulting in the
smallest delays.
Here
is the condition for setup timing check; which does not include any of the OCV
settings for the derated delays [3].
Launch Clock Path + Max Data Path < = Clock
Period + Capture Clock Path Tsetup_UFF1 [3]
It implies that the minimum clock period =
[Launch Clock Path + Max Data
Path – Capture Clock Path -Tsetup_UFF1]
From the Fig. 1.1, without considering effect of OCV,
Max Data Path = 5.2
Launch Clock Path = 1.2 + 0.8 = 2.0
Capture Clock Path = 1.2 + 0.86 = 2.06
Tsetup_UFF1 = 0.35
Resulting in a minimum clock period of:
2.0
+ 5.2 – 2.06 + 0.35 = 5.49ns
Using
the commands related to setting of timing derate, cell delays can be
derated.
So,
for example it can be supposed that we are derating the early/minimum/shortest
paths by -10% and derating the latest/maximum/longest/ paths by +20%.
The
derate values specified with the -late option for the command to apply
derate , is multiplied with the delays of long paths such as launch clock path
and data paths for setup checks or capture clock paths for hold checks.
The
derate values specified with the - early option for the command to apply
derate, is multiplied with the delays of short paths such as launch clock path
and data paths for hold checks or capture clock paths for setup checks.
A
value of 1.0 is assumed for derating factor, if no derating factors are
specified.
The
derating factors get applied uniformly to all the cell delays and the net
delays.
The
-net_delay and the -cell_delay the options can be used with the command used to (set the timing derate) at a
scenario-based application where different derating factors needed for the nets
and the cells.
The -data and
the -clock option can also be used
with the command used to set the timing derate to apply derating only to data
and clock paths respectively.
We now try to apply derating derating to
the early/minimum/shortest paths by -10% and derating the latest/maximum/longest/
paths by +20% and also apply late derating of 10% to cells also , to the
example of Figure 1.1 [3].
With these above specified derating values, we get
the following conditions and calculations for setup check as in next page.
Max Data Path = 5.2 * 1.2 = 6.24
Launch Clock Path = 2.0 * 1.2 = 2.4
Capture Clock Path = 2.06 * 0.9 = 1.854
Tsetup_UFF1
= 0.35 * 1.1 = 0.385 ( here we have used
10% late derating for the cell, so a factor of 1.1 needs to be multiplied to
the setup of this flop)
In
the above setup check, there is a discrepancy with a delay of 1.2ns, as it is
being derated differently for the capture clock and for the launch
clock, due to the common clock path (Fig. 1.1) of the clock tree. The pessimism caused due to
application of different derating factors on the common part of the clock tree
is called Common Path Pessimism (CPP) which should not be there during
the analysis [3].
Artificially
induced pessimism removal between the capture clock path and the launch clock
path during timing analysis is Common Path Pessimism Removal (CPPR)
which is also known as Clock Reconvergence Pessimism Removal (CRPR). And
is often listed in the path report as a separate item [3].
CPP = LatestArrivalTime@CommonPoint –
EarliestArrivalTime@CommonPoint
LatestArrivalTime@CommonPoint = 1.2 * 1.2 = 1.44
EarliestArrivalTime@CommonPoint = 1.2 * 0.9 = 1.08
Which results in a CPP of: 1.44 - 1.08 = 0.36ns
Including the CPP correction, it results in a,
Minimum clock period of: 7.171 - 0.36 = 6.811ns
The maximum
frequency at which the design is operating can be reduced by use of these
derating factors to model the OCV variations.
1.1.4 ADVANCED ON-CHIP VARIATIONS (AOCV)
To account for OCV,
there have been a conventional way for many years to apply global, a single
derating value to the design to add extra design margin. For the process
variations for 90 nm and 130 nm design, this conventional approach is
reasonable to apply. But for designs at smaller geometries, it may add
excessive pessimistic margins to the design and hence result in over design,
reduce performance of the design and also result in longer design cycles.
Advanced on-chip
variation is naturally exhibited from
OCV which is an improved way of introducing margins in the design, it
models systematic and random variations which effect timing across an IC by
making use of variable derating factors based on the depth of every path and
location being analysed [4].
Instead of using
single global value for derating, using context specific value for derating, we
may reduce extra design margin and hence have lesser timing violations.
AOCV which is
approach to reduce pessimism, reduces unwanted over-design and helps us to
achieve timing closure sooner and with reasonable confidence.
AOCV related information is specified using derating value tables.
Variable value derating factors are calculated and applied based on the depth
and location of every path being analyzed.
AOCV is summed into
two the path based and group based analysis processes, which enables successive
policy of optimization for analysis that’s used with conventional Static timing
analysis already today.
AOCV Path based and
Graph based analysis is analogous to Path based and Graph based analysis in the
conventional static timing analysis context [4].
1.1.4.1 AOCV GRAPH BASED ANALYSIS
·
Making
use of update_timing command, it delivers design-wide, fast analysis.
Using AOCV instead of conventional
OCV, helps in reducing the pessimism.
·
AOCV
Graph based analysis for most of the designs is signoff sufficient.
·
For
margin reduction and further analysis AOCV graph based analysis is a necessary
prerequisite using path based AOCV analysis on critical paths selectively.
1.1.4.2 AOCV PATH BASED ANALYSIS
·
Paths
which are critical and still violating after performing Graph based analysis,
path specific analysis is recalculated to reduce the pessimism further.
·
To
determine derating values that are path specific, it uses location based
bounding box and path specific path depth values.
.
The cell path depths and the net
path depths might be different. Fig. 1.2, shows an illustration to calculate
the cell path depth.
By the measurement of the diagonal of a box that encompasses all the cells in the path as shown in Fig. 1.3,
the distance of the cell path is calculated.
In
the same way, by the measurement of the diagonal of a box that encompasses
all of the nets in the path as shown in Fig. 1.4,
the net path distance is calculated.
Because the capture and the
launch signals in the common path which passes through the common point, are
the same signals. Therefore the tool assumes, zero variation in the common path.
So, for AOCV metric computations Tool by default, excludes cell or net objects
in the common path.
Figure 1.2: Path-Based Cell Depth
Metric Computation
Figure 1.3 Path-Based Cell
Bounding Box Distance Computation
Figure 1.4: Path-Based Net
Bounding Box Distance Computation
·
A
path’s slack is not worsen by this AOCV Path based analysis
Logic depth and location attack two different parts
of the problem.
Random Effects (depth)
- According
to Synopsys, statistical analysis shows that deeper paths are less
affected by random variation - because the contributory effects are
random, all cells within a deep path are highly unlikely to be
simultaneously fast or slow.
- CLT
(Central Limit Theorem) establishes that, in most situations,
when independent random variables are added, their normalized
sum tends toward a normal distribution as the number of values tend
to be sufficiently large.
- Since the
normal or Gaussian curve is symmetric about the mean, we can say that the
random effects cancel each other and hence as the depth increases,
derating factor decreases.
Systematic Effects
(distance)
- Location-based
derating deals largely with systematic effects. Cells in close proximity
exhibit less variation relative to one another than those further apart.
- Using
silicon data from test-chips, derate factors can be calculated based on
relative cell-location. Advanced OCV computes the length of the diagonal
of the bounding box that contains the cells being analysed to select an
appropriate derate value from the table constructed by test-chip results.
Graph based analysis must be performed before path
based analysis generally as path based analysis is computationally more
intensive. With OCV and conventional Static timing analysis, path based
analysis increases or scales with number of paths for analysis and may have
exponential runtime [4].
Figure 1.5: Pessimism Reduction
by OCV Modelling Methodology
SOCV AND POCV ARE IN NEXT BLOG CONTINUED
Comments
Post a Comment