Design for Test [DFT]-1 (1).pdf DESIGN DFT

Design for Test [DFT]
• System on Chip (SOC): All the blocks like memory, ALU, RAM,ROM,
Microprocessor, Microcontroller etc are integrated on a chip is called
as SOC. It mainly consists of 4 blocks.

Design for Test [DFT]
• DFT: Generates the test patterns and those patterns are able to
detect any manufacturing defects.
• DFT engineers job is to facilitate the mechanism to identify the
defective parts & non defective parts.
• It provides interfacing between pre-manufacturing & post-
manufacturing.
DFT Engineers – Gives the Test vectors to the Test engineers.
Test Engineers – take the test vectors and applies to the chip,
specifies chip pass/fail.

Goal of DFT:
Goal of DFT is to identifying the defects and separating the faulty parts.
But not identifying where is the defect, how the defect occurs.
• DFT Engineers provides i/p’s in the format of Test vectors to the Test
Engineer team to apply on the manufactured IC and prove the IC’s
pass/fail.
• To test the Digital type logic, we use Scan method .
• For I/O defects we use – BSCAN (Boundary Scan)
• For memory – MBIST (Memory BIST)
• For Analog – ADFT (Analog DFT)
There is no Generic way to Test Analog (There is no fixed methodology
to test Analog).

Design Structures:
1) Digital: (i) Combinational Logic
(ii) Sequential Logic
2) Inputs & Outputs (I/O)
3) Memories
4) Analog

Inputs & Outputs:
It takes input from the external input world and gives to the input pad,
then to some internal logic and further sends the output to external
world through output pad.

Bidirectional Pad:
• It has a logic inside the chip,
If oe=1, then pad works as an output pad.
If oe=0, then pad works as an input pad, i.e., signal going into chip.

Tri - State Pad:
When oe=1, the output goes to the external world.
When oe=0, it creates an high impendence state, i.e., it completely
block signals going to i/p (or) o/p.

IC Classification:
S.no No of Transistors per Chip Technology name
1 Less than 100 SSI
2 100 – 10,000 MSI
3 10,000 – 1,00,000 LSI
4 Greater than 1,00,000 VLSI
5 Greater than 100 millions ULSI

Design for Test [DFT]-1 (1).pdf DESIGN DFT

• Library is a file, which we have all combinational and sequence
elements.
• If OR is there in netlist, tool won’t understand it, it will check in the
library for OR gate, if it finds there then it will perform the
corresponding operation.
• Eg: OR2x4_LVT
2 – no of i/p’s
X4 – Drive strength – capacity of the cell to drive the load.
(Drive Strength means Fan-Out)number of gates that an output of a
gate can drive.

Actually 3 types of library cells
LVT (Low Threshold Voltage)
RVT (Regular Threshold Voltage)
or
SVT (Standard Threshold Voltage)
HVT (High Threshold Voltage)
1) Fast cell
2) High Leakage
1) Medium Cell
2) Medium Leakage
1) Slow Cell
2) Low Leakage

Cont..
• Suppose if we want to build a processor that works with higher speed
in a Giga Hz, then we have to pick OR2 cell with LVT.
• Similarly, if we want low power devices then pick OR2 with HVT.
• If it is a Trade off then we go with RVT.

• Library means not only AND, OR, NOT, XOR, XNOR, it also has drive
strength, and drive strength has different thresholds (LVT, RVT, HVT).
Eg: AND2_X4_LVT
RTL code Netlist
• Maximum drive strength is 32,it will change depending upon
technology
• Tool reads the RTL, understands the functionality and it will converts
into gates from the library and tries to implement same functionality
coded in RTL.
Synthesis
Tool (Dc Compiler)

1. RTL Functionality = netlist functionality.
2. It also do optimization (PPA)
a) Power b)Performance c) Area
i.e., Power Optimization go for HVT cell.
Performance Optimization go for LVT cell.
Area Optimization drive strength.
(if drive strength is more, then more gates are there, then size (or) area
increases)
 we prepare to take lower drive strength.

• Suppose, if we need 100MHz speed but with lesser area
• Suppose, this will be done with X2 and X4, then tool picks X2 only.
 depending on design constraints given to the synthesis tool, it will
generate a netlist.
• Suppose, if we want the chip to work at 500MHz, we cannot get
500MHz from the entire design. Some blocks has 500MHz, some has
50MHz, 100MHz, 250MHz….
 The tool will decide, whether to use LVT, HVT,RVT (or) combination
of LVT and HVT depending on PPA (Power, Performance and Area).
i.e., 20% of LVT, 60% of LVT + HVT, 20% HVT that will be decided by the
tool depending on constraints.

ASIC FLOW
• IC for a specific functionality, eg: mobile, laptop.
• Same design is used to manufacture millions of products.
• When we need higher volumes of IC’s we go by ASIC.
FPGA:
• It supports different applications( like adder, subtractor, multiplier..).
i.e., Reconfigurable.
• Whereas ASIC is application specific IC, i.e.,suppose an adder ASIC only performs
Addition.
• Suppose Adder is designed with ASIC & FPGA.
• Then PPA is better in ASIC then FPGA.
• ASIC performs only one application, whereas FPGA is reconfigurable hence extra
circuit is needed for FPGA.

Defects in Silicon
• IC manufacturing process is defect-prone.
• Database given for manufacturing and outcome of a manufactured chip both are functionally equal.
• I need to prove that, they are equal or not.
If they are equal it is fault free
otherwise , there is a defect or fault.
• Defects during manufacturing open or short are caused by impurities, shorts for additive material,
opens for subtraction material.

Testing:
• Testing is a process of sorting DUT’s to determine their acceptability.
• If we apply 100 patterns as stimulus it should
#1 pattern what is the o/p
.
.
.

• First we should have expected o/p’s
• Then apply stimulus (I/P pattern)
Suppose
I/P O/P
#1 1
#2 0
#3
.
.
0
.
.
#100 1
I/P O/P
#1 1
#2 0
#3
.
.
0
.
.
#100 1
if any one pattern fails then it is a defective chip and it
is separated.

Quality – Defective Parts Per Million (DPPM)
Yield:
• Suppose 100 IC’s are manufactured and all 100 IC’s are pass then we can
say 100% “Yield”.
• Suppose 90 IC’s are pass then we can say 90% yield; 10% is yield loss.
 Yield =
• i.e., How many pass, after manufacturing is called “Yield” .
DPPM:
• Suppose 1 Million IC’s are shipped to the customer, after DFT, suppose 100
IC’s are returned back from the customer,
then DPPM = 100.
Note: Even after DFT, there may be possibility of chip fail.
How many pass
Total Manufactured

ATE (Automatic Test Equipment)
• After manufacturing we are using ATE to test.
TE (Test Engineer):
• Work is writing test program, applying them; and during testing if the
chips are failing, then he go back to DFT Engineer.

Multisite testing
1) One ATE tests several (usually identical) devices at the same time.
2) For both probe (wafer level testing) and package test(taking a die
from wafer, and putting in a package).
3) DUT interface board has >1 sockets.
4) Add more instruments to ATE to handle multiple devices
simultaneously.
5) Usually test 2 or 4 DUT’s at a time, usually test 32 or 64 memory
chips at a time.

Cont…
• Suppose, if we have 100 million IC’s,
• Let us assume it will take 1min to test one IC, then to test 100 millions, it
will take 100 million minutes.
• Suppose, a mobile has 3 IC’s , for 100 million mobiles, it will take 30 million
minutes.
• Therefore, to increase the testing speed, instead of testing one IC, you can
test 2 IC’s at a time is called Dual Site Testing
• 4 IC’s at a time – Quad site testing.
• 8 IC’s at a time – Octal site testing.
, if we test 8 IC’s in a min then we need
= 12 million minutes instead of 100 million minutes.
100 millions
8

Cont….
• Here, ATE generates test pattern to the wafer, then failure dies are
identified , by taking the information of fail dies from ATE, the wafer
map will Ink the dies which are fail.
• Now by using dicing and binning , here sorting is done, i.e., good parts
(dies) are Dicing (cut and send to package) and bad (or) Ink dies are
moved to Bin.

Manufacturing Test:
1) A speck of dust on a wafer is sufficient to kill chip.
2) Yield of any chip is <100%.
Must test chips after manufacturing before delivering to
customers to only ship good parts.
3) Manufacturing testers are very expensive
a. Minimize time on tester
b. Careful selection of test vectors.

What is DFT ?
• It is a technique of adding testability features to IC design.
• Developing and applying manufacturing tests.
• The manufacturing tests help to separate defective components from
the healthy ones.
• Actually, let us assume an AND gate IC.
• During the manufacturing , test logic also inserted.
• Now apply test pattern to Test logic, to check the functionality of AND
gate, After verification, if everything is ok then manufacturer disables
the Test logic.

What are DFT Goals ?
Quality: A high degree of confidence during testing
Test cost reduction
Time to volume reduction through test automation.
Ad-Hoc DFT:
To enhance a designs testability without major changes to the design style.
Minimizing Redundant logic
Minimizing Asynchronous logic
Isolating clocks from the logic
Adding internal control and observation points.
However structured DFT techniques yields greater results.

Structured DFT:
1) More systematic and automatic approach.
2) Scan Design
3) BIST
4) Boundary Scan
• Goal is to increase the Controllability & Observability of a circuit.
• Actually by using Ad-Hoc DFT we cannot test each and every block
but by using structured DFT we can test each and every block.
like, to test digital logic – use Scan methodology
to test memories – MBIST
to test I/O’s - JTAG (or) Boundary Scan.

• Suppose, if we want to manufacture an AND gate,
• Here 3 nodes are present, during manufacturing there are
two possibilities
1) stuck at – 0;
2) stuck at – 1
• Each node has a possibility of 2 defects.

Therefore, a simple AND gate with 3 nodes has 6 possible defects.

• DFT helps to
a) Verify chip’s functionality.
b) Reduce debug time.
c) Faster ramp to volume production.
o/p: b & c.

Definitions:
1) Defects :Imperfection of flow that happen in a particular DUT.
2) Faults : A representation/modelling of that defect.
3) Failure: Non- Performance of the intended function of the device
mainly due to the defect.

DFT Definitions:
A Design is Testable:
• All internal nodes of interest are simultaneously controllable ( to a
desired logic value) and observable.
Design for testability:
• A methodology (or) a collection of methodologies, which results in
the creation of a testable design.
Diagnosis:
• To locate the cause of misbehavior after the incorrect behavior is
detected.

Functionality vs Structural ATPG:

Cont…
, 13x2 =26
, for 64-bit, total no of faults = 26x64 = 1664
,for 64-bit adder, we have 1664+640=2304 faults.

Functional vs Structural testing
• If we go by functional way of testing
• Functional ATPG – generate complete set of tests for circuit i/p – o/p
combinations.
129 – i/p’s , 65 – o/p’s.
, 2129 = 680,564,733,841,926,749,214,863,536,422,912 patterns.
by using 1GHz ATE (High speed ATE), would take 2.15x1022 years.
(which is not feasible).

Cont….
• Where as by structural test,
No redundant adder hardware, 64 bit slices.
Each with 36 faults (using fault equivalence)
At most 64x36=2304 faults.
By using 1GHz ATE – 0.000001728 sec (1.7μs).
 Industries prefer structural testing.
suppose, with structural testing we got 99.5%.
• For coverage of 0.5% go for functional testing.
designers gives small set of functional tests – augment with
structural tests to boost coverage to 98+%.

Design Environment Setup
Link library
Target library
Search path
Tool Invocation
Netlist
Constraints

• Suppose, the tool does not find latch cell in link library then it shows
error like Black Box (or) cell not found etc..
• Eg: I have 40nm synthesis, I want to convert this in 28nm netlist, to do
Scan Insertion.
i/p is 40nm netlist we need link library (read).
Now o/p is 28nm netlist we need Target library (write).
Link Library:
• The tool uses the link lib to read i/p netlist (reading all the cells) (or)
for each cell (like OR, AND , NOT, LATCH…) the tool will go and look
into the link library.

Cont…
Target Lib :
• It is related to the tool writing the o/p netlist depending up on the
constraints (or) requirements like area optimization, power optimization,
performance optimization etc… the tool write the netlist by taking all the
cells from the Target library.
Eg: (I/p) (o/p)
50 MHz with 40nm to 100 MHz with 40nm
Here technology is not changed ie.,both are in 40nm.
Link library is same as Target library but to get 100MHz we have to
increase the performance then change the cells like LVT,HVT,RVT.
Synthesis will pick the correct cell to improve the performance to the
required level.
Link Library Target library (or) link Library

• Eg: 100MHz with 40nm 100 MHz with 28nm
Search Path:
• It is the path where you will get all libraries and telling the tool where
to look for Target lib.
To read i/p link lib
 Technology is different
 Target lib is used to write the o/p netlist

Single Struck-at fault:
• Only 1 node is faulty in the entire design at one point of time, it is
permanently S@0 (or) S@1. it does not toggle.
• The fault can be either at i/p (or) o/p of a gate.

• Here, observability failure is gone, because , we can put 1/0 at that
node and that 1/0 can be transferred to PO(Primary O/p).
• Similarly Controllability failure is gone, because we can put 0 at PI
(Primary Input) and by setting Test_mode =1 then we can get 0 at
o/p of MUX ( at any node we can put 1/0)
If Test_mode=0 then we are enabling the OR gate (functional logic)
through OR gate take place.
If Test_mode=1, then we are disabling the OR gate.
 here for one OR gate, we have used, 1 Mux ,1 PO,1 PI.
• Suppose for 100 OR gates we have to consider 100 Mux, 100 PI, 100
Po.
• Theoretically, it solves the problem but practically it is not possible to
take “n” no. of i/p’s.

(i) Controllability:
The ability to set a node in a design to a desired state, i.e., logic 0 (or) 1.
(ii) Observability:
The ability to observe a change (valid o/p) in logic value of a node in a
design.
(iii) Testability:
If a design is well-controllable & observable it is said to be easily
testable.
(iv) Not possible to add such directly controllable & observable points
to every gates in an actual design.

To avoid this, scan design comes. (to avoid using PI & PO).
Actually any digital path you take it will fall under 4 paths. i.e.,
I. I/p – Reg
II. Reg - Reg
III. Reg – O/p
IV. I/p – O/p

Cont….
• But how can we test the combination circuit which is inside the chip,
i.e., combinational circuit sitting between I/P and Reg (f/f) (or)
between Reg to Reg (or) between Reg to O/p.

Cont…
• At this station , we use virtual i/p from the reg and virtual o/p from the reg.
• That means we have to use Scan Design, to controllability & observability of the
logic sitting internally in the chip.

Cont…
• We cannot access the D-ff , which is internally sitting in the chip.
So we placed a 2:1 MUX before a D-ff during manufacturing.
If scan_en =0, then normal D-ff operation takes place (functional
data transferred into the D-ff it is called functional path).
If scan_en =1, then scan_data (either 1/0) is transferred to D-ff by
applying sys_clk that 1/0 is stored in D-ff, we can access through Q.
 by doing this controllability & observability are there.

Cont….
Before scan insertion After scan insertion

Scan Operation:
1) Select shift mode, SE=1.
2) Shift-in/load scan cell values.
3) Select capture mode, SE=0.
4) Apply/force primary i/p’s.
5) Measure primary o/p’s.
6) Capture combinational logic response into scan flops.
7) Select shift mode, SE=1.
8) Shift-out / unload the scan cells data.
9) Shift-in the next scan pattern.
Process repeats again and again.

Scan chain operation for stuck at Test:

Scan Test Time:
• Suppose , If we have 1000 flops.

Cont…
• The optimization is, during 1st pattern shift out we require 1000
pulses, at the time of shifting suppose if we use 2nd pattern at scan in
for shift in (load) and 1st pattern shift out are taken care by only 1000
pulses.
• In this way we can reduce the Test clock cycles (or) Test time.
Most of the clock cycles are consumed by shift in & shift out.
• Shift in & shift out are depends on scan chain length.
, if we are having shorter chains then Test time will be reduced.

Scan Benefits:
1) Makes the complete design look like a shift register.
2) Scan enables us to get data in (to the design) and data out (of the
design) easily.
3) Easy test pattern generation for design.
4) Basis for many ATPG related tools.
• Which comes first in the SoC flow?
a. ATPG
b. Pattern simulation
c. Scan Insertion
d. Pattern Conversion Ans:c,a,b,d.

Scan Models
• Most commonly used scan models are:
MUX-DFF
Clocked Scan
LSSD (Level Sensitive Scan Design) (Latch based)

Clocked Scan
• Actually clk itself consumes more power
• Here we are using another clk i.e., SC_Clk,  increases the complexity and power
requirements.
• Uses dedicated test clk (SC-CLK) in addition to sys_clk.
• Operation:
Functional mode: Data, sys_clk
Test mode: SC_IN, SC_CLK
• Need to route an extra clock tree.

LSSD:
• Uses 2 test clocks: Aclk, Bclk.
• Operation:
Functional mode: data, sys_clk.
Test mode: sc_in, Aclk, Bclk.
• Need to route an extra clock tree.
• Additional area due to extra location.

Comparison of Scan Designs
Scan Design Advantages Disadvantages
MUX-D Scan Supports all automation tools Mux delay
Clocked scan No performance degradation Addition shift clock routing
LSSD No race condition Area & Routing complexity

Scan Operation
Clock
Any signal that can change the state of flipflop (or) Reg is called clk. (set (or) reset are
also called clock).

Scan Golden Rules (or) DRC’s: All DRC’s are netlist based.
DFT rule 1:
• All the internal clocks must be controlled by port level clock signal (primary i/p) in scan test mode.
• The internal pin Q may be 1 or 0. hence we cannot control the clock of FF2.
• Actually all the internal clocks must be controlled by the port level clk, for this to happen, a mux is
placed, this mux should not effect the functionality (i.e: Q to clock connection).

DFT Rule: 2
• Avoid implementation of combinational feedback circuit.
If present, the feedback loop will be broken to test.
• The feedback signal may not be testable (observable) in test mode.

DFT Rule: 3
• Asynchronous SET/RESET pins of flipflops must be controlled by a port
level RESET (primary i/p) in Scan Test mode.

DFT Rule: 4
• Gated clock must be enabled in scan test mode.
• Whenever clk is applied at port level, that will reach as i/p to
AND gate; if 2nd i/p to the AND gate is ‘1’ then clk will be propagated;
if 2nd i/p is ‘0’ then clock won’t propagate.

Cont:
• If the Testmode=1, o/p of AND is 0, then latch will store this ‘0’ and at
o/p of latch it is ‘1’.
, clk is propagated, (we need D=0 always).
• But here there is no controllability of Hold.
i.e: if the Hold receives the i/p from any combo logic then additional
scan flop is needed to observe the value.
• If the Hold receives i/p from any scan flop then the additional scan
flop is not needed for observability of Hold value.

DFT Rule: 5
• Latches have to be avoided as much as possible, if present, make it
transparent in scan test mode.
• In an Edge- triggered design, it is difficult to put latches on a scan
chain because the library does not contain their edge- triggered scan
equivalents.
• If they (latch) are not part of a scan chain, their o/p’s will be difficult
to control. The faults coverage will therefore be very low.

DFT Rule: 6
• Do not replace flipflops of the shift register structure by equivalent
scan flops.
• i.e., suppose , if we are not having combinational logic to test, then
no need of scan flops.
• Similarly, if we are having shift registers in a functional block, and
there is no combo logic in the functional block, then there is no need
to convert the flipflops of shift registers to scan chain.

• Similarly if there is a combo logic, and a 32-bit shift reg in the functional
block, then we need to change only one (first) flop in the shift reg to scan
flop.
• Then to test combo logic, only 0th flop needs to convert as scan flop and
remaining 31 flops will remain as normal shift-reg.
• For efficient area purpose, the flipflops of the shift register structure will
not be replaced by equivalent scan flipflops.
• Compare to normal flops, scan flops takes more area.

DFT Rule: 7
• Clock should not be used as data in scan test mode.
• If Testmode =1; then o/p of OR is 1.
• Now, i/p to AND is 1 and another i/p is from Data,
•  if data =10101, then that data will be stored in MUX Scan flop.

DFT Rule: 8
• By pass the memory in scan test mode.
All the paths ending at memory cell are not observable.
All the paths starting from memory are not controllable.

Suppose, a combo logic is present like this,
• The o/p of the combo logic is hitting the memory, if the ,memory is present in the functional design then it
is like dead end, we cannot observe the value.
 for observability we bypasses to mux through a flop.
• Similarly, the o/p from memory is also not controllable.
• In the bottom, a combo logic is present, for observability a flop is placed.

DFT Rule: 9
The scan enable signal must be buffered adequately.
1) The scan enable signal that causes all flip flops in the design to be
connected to form the scan shift register, has to be fed to all flip
flops in the design. This signal will be heavily loaded.
2) The problem of buffering this signal is identical to that of clock
buffering.
3) The drive strength of scan enable port on each block of the design
must be set to a realistic value when the design is synthesized.
4) If this port is left unconstrained during synthesis, it could result in
silicon failure.

DFT Rule: 10
• Avoid multicycle paths as much as possible .
(Ideally Zero)

DFT Rule: 11
• Negative edge flops should be placed in the start of the scan chain.

See in previous figure for 4 clk pulses, the value loaded is 1101 but not
1010. this is because –ve edge flops are taken in the middle. Those
must be at the beginning.

Scan Types
1) Full scan
2) Partial scan
3) Partition scan

(1)Full Scan:
Full scan is a scan design methodology that replaces all memory elements in
the design with their scan-able equivalents and then stitches them into scan
chains.
Benefits:
1) Highly –automated process(tool generates less pattern for best coverage)
2) Highly – effective, predictable method.
3) Easy to use.
4) Assured quality.

(2) Partial Scan:
Full scan design makes all storage elements scannable, it may not be acceptable for
all your designs because of area and timing constraints.
Where as partial scan is a scan design methodology only a percentage of the
storage elements in the design are replaced by their scannable equivalents and
stitched into scan chains.
Benefits:
1) Reduce impact in area
2) Reduced impact on timing
3) More flexibility between overhead and fault coverage.

(3) Partition scan:
• It is for complex SoC designs.
• The ATPG process on very large, complex designs can often be
unpredictable.
• Large designs which are split into a no. of design blocks.
Benefits :
1) Improves test coverage and runtime.
2) Less power consuming since each block tested separately.

How to fix below DRC issue?
• We cannot control the clk, because o/p from combo logic is ‘0’ then
the clk won’t propagate to the flop. So o/p of combo logic maybe 0
(or) 1. the clk should be from top level only.
we modify this circuit.

• Combinational Feedback loop?
• We cannot observe the o/p Q because it keeps on changing for
observability we have to break that loop by placing OR gate.

Lockup Latches
• Block A is following the guidelines (placing –ve edge clk at first place)
similarly Block B also following the guidelines.
• But when both blocks are combined then they violated the rule, at
this situation lockup latches are used.

• Even though we are giving clk pulses to all the flops, there is a clock skew
(i.e., expected-arrival) i.e., the 1st flop receives clk first, and further flop
receives last because of this,clk skew is present, at this situation lockup
latch is introduced.
• The lockup latch is active low enable,
• It holds the data for half cycle.
• In the industries, lockup latches are inserted at the end of the every scan
chain within the block itself.

Clock Gating:
• A latch is placed, to reduce the glitches.
• En(enable) acts as data pin, here clk is –ve edge (or) –ve level triggered.
• The target is to get neat gated clk at o/p.
• A latch with a gate is called ICG (Integrated Clock Gate) it comes as one macro cell. i.e., It
is not like one latch and one separate AND gate combined.

Now we know if clk2 is coming early than clk1 then data loaded
correctly, what if we inserted a lockup latch here,

• When two clk domains are
combinely used in a chain
then we cannot see which clk
is coming first and which one
is second.
• If clk2 comes 1st than clk1
then no issues, if clk1 comes
1st than clk2 then there is
issue.
• So, at that situation lockup
latches are used , to load the
data correctly.

Design for Test [DFT]-1 (1).pdf DESIGN DFT

More Related Content

What's hot (20)

Similar to Design for Test [DFT]-1 (1).pdf DESIGN DFT (20)

Recently uploaded (20)

Design for Test [DFT]-1 (1).pdf DESIGN DFT