1. Bill Verzi Automated Test Group February 24, 2005
Common Mistakes of Parametric Test…
… and how to avoid them!
Tutorial lecture
presented at the
IEEE Electron Device Society
Central Texas Section
Austin, Texas
February 24, 2005.
The Most Common Mistakes Made in Parametric Test

2. Outline of Lecture Introduce Semiconductor Manufacturing and Test
Using electrical test to evaluate semiconductor process features
Top Ten List of Common Mistakes, and how to avoid them!
The Most Common Mistakes Made in Parametric Test

3. An Introduction to D.C. Parametric Test Engineering
Section 1
An Introduction to
D.C. Parametric Test Engineering
The Most Common Mistakes Made in Parametric Test

4. Why “Parametric” Test MOSFET Id equation Capacitance Resistivity
I am often asked, “Why are you called a D.C. Parametric Test Engineer?” The answer provides insight into the nature of our work. Although we are test engineers, we do not test products, but rather, the semiconductor manufacturing process. Semiconductor device activity can be described mathematically, with variables, or parameters as components of each equation. For example:
MOSFET Id equation Id=Z/L UsCox(Vg-Vt)Vd
Capacitance C=k A/d
Resistivity R= Rho L/W
Each equation is composed of constants and variables. Certain variables are held constant while others are measured to complete the equation. These variables are parameters of the equation.
Thus, we perform parametric test to solve for parameters of equations that describe the functionality of solid state devices. There are hundreds of equations that can be formed to completely describe semiconductor devices. Each equation motivates a DC Parametric measurement that will be used to evaluate, characterize, or control the manufacture of the device.
The Most Common Mistakes Made in Parametric Test

5. The big secret! This is the easiest test there is...
Ohm’s Law
R=V/I I=V/R V=IR
Test Structures are variants of:
Diodes
Resistors
Capacitors
Transistors
But we measure these devices very carefully!
Although many other forms of product-level test are quite complex in development and understanding, D.C. Parametric test is relatively simple to derive. In most cases, the measurements we make can be simplified to a measurement of one of the following devices; Resistors, diodes, capacitors, or transistors.
All of these devices are evaluated by providing a D.C. stimulus in current or voltage, then measuring a response in current or voltage. In the special case of capacitance measurements, we apply an A.C. stimulus, but we do so only to measure the complex impedance of the capacitor, returning us to simple resistance measurements.
The complexity of D.C. Parametric test comes from the quality of measurement. We measure in units of: FemtoAmperes, MicroVolts, and PicoFarads. These measurements require careful measurement techniques on structures specially designed to produce minute yet meaningful results.
Given the hundreds of parametric equations that describe the semiconductor device and manufacturing process, and considering the hundreds of individual physical or electrical parameters we wish to control, we design test structures in such a way as to isolate and emphasize these features. Special structures are designed to isolate individual physical and electrical process artifacts so that each feature may be measured and understood independently from other artifacts of the process.
Test structures are created to evaluate and control the ‘Design Rules’ of the process, Random and Systematic Defects in the process, Performance Metrics of devices created from the process, and the assessment of the life-time or Reliability of the process.
The Most Common Mistakes Made in Parametric Test

6. Environment where DC Parametric Test Systems are Used
Clean Room
Bunny suits
Automatic
Wafer Prober
Darn good DC Parametric Test System
The Most Common Mistakes Made in Parametric Test

7. Semiconductor Wafers in Laboratories
What do we measure?
Semiconductor Wafers in Laboratories
The Most Common Mistakes Made in Parametric Test

8. Semiconductor Wafers in Production
What do we measure?
Semiconductor Wafers in Production
Scribe Line TEG
Scribe line
Pad
Minor flat
In-scribe test
pattern
Test Die
Die or Chip
xxxx-xxxx-xx-x
Wafer ID
Alignment
target
Major flat
The Most Common Mistakes Made in Parametric Test

9. A CMOS Cross-Section
This cross-sectional representation of a modern CMOS technology demonstrates the many features, dimensions, and artifacts of the manufacturing process that can be assessed through D.C. Parametric test.
The Most Common Mistakes Made in Parametric Test

10. DC Parametric Test for Front End Defects, Design Rules, Performance, and Reliability
Metal 1 contact
resistance,
1st insulator bridge and space
Tungsten and
interface
resistance,
Contact size and
alignment
Contact to gate space design rule
Front End tests and measurements include:
Defect testing: Oxide breakdown
Polysilicon bridging
Design Rule testing: Critical Dimension evaluation such as
Poly Width
Isolation width
Layer-to-layer alignment
Contact to Gate Space
Contact size
Thick field threshold
Latch-up effect
Performance Metrics:
MOSFET parameters: Vt, Ids, Gm, Rds, Peak Isub, Vpt, Bvdss, …
Reliability Tests:
For Oxides: TDDB, GOI, Bvg
For MOSFETs: HCI, Fluence, Peak Isub, Vbii
Poly and silicide
resistance,
Junction spiking
Transistor Vt, Gm,
Rds, Ids, Vpt,
Peak Isub, Bvdss
Thick Field Vt
Mobile Ion,
Latch-up
Oxide Integrity, Qbd,
Bulk impurities, TDDB,
Well capacitance and resistance
Junction leakage
and Breakdown
The Most Common Mistakes Made in Parametric Test

11. DC Parametric Test for Back End Defects, Design Rules, and Reliability
Insulation breakdown, Back-end contamination tests,
Intra and Inter-Layer bridging and space design rules
Metal resistance,
Width and space
design rules,
Electromigration
Contact size and
alignment,
Contact and
interface
resistance
Metal width and
space design rules,
Metal bridging and
continuity
Back End tests include:
Defect testing: Inter layer Dielectric Breakdown
Layer to Layer bridging and Inter-layer bridging
Contact defect density
Design Rule testing: Critical Dimension evaluation such as
Metal Width
Isolation width
Layer-to-layer alignment
Contact to Edge Space
Contact size
Thick field threshold
Reliability Tests:
For metals: Electromigration, Isothermal and SWEAT
The Most Common Mistakes Made in Parametric Test

12. Statistical Process Control and Parametric Test
All of these structures, all of these methods …
and the wafers are very large!
Plus my cycle times are not even close to 90 days
I’m lucky to get 3 information turns per year!
How can I handle all of this data to be sure I get it right?
Statistical Process Control is the means by which we handle this job!
For modern technologies there may be hundreds of individual features. In R&D, the number of structures can be three to ten times the number typically used in production. This is because in R&D the design rules for the process are still being evaluated. Three to ten similar structures may be drawn with the only difference in design being small increments of dimension for the isolated feature. As an example, the size of contact may be set at .2, .225, .25, .275, or .3 microns. Experiments are executed to identify the smallest contact size that produces the resistance and yield needed for functional products to be manufactured. Therefore, there are 5 structures needed for this single process feature. This pattern of structure design is repeated for each of the hundreds of process features or artifacts.
Additionally, the variance of a process is also related to the position of the feature on the wafer. With wafer diameters exceeding 200mm, the spatial variation of the process can only be evaluated by measuring more locations on the wafer. This increases the time to test the wafer.
If my cycle times are not even close to 90 days, but more like 120 days, I’m lucky to get 3 information turns per year! I can’t afford these difficulties that increase the time of the information turn and introduce significant complexity in the analysis of the process.
How do you handle all of this data to be sure you get it right? Statistical Process Control is the means by which we handle this job!
The Most Common Mistakes Made in Parametric Test

13. What is Statistical Process Control?
Parameter statistical samples
( Mean, Sigma, Range, … )
By measuring samples of the population of the controlled parameter, an understanding of the character of the process is gained. Each individual process parameter is measured with enough samples to achieve a confident understanding of the process and spatial components of variance of the parameter.
At the beginning of the technology development cycle, each parameter is uncontrolled. A parameter is out of control when the mean of the sample is different from the target specification for the parameter. A parameter can also be out of control when the sigma or range of the parameter varies greatly over time.
But with each information turn, the variance of the parameter is reduced until process parameter control is achieved. Process control is achieved when the mean of the parameter is acceptably near the target specification for the parameter and the variance of the parameter is consistent. Over time, repeated measurements indicate a trend that can be used to control the manufacturing process.
Process engineers use methods similar to those described above to control the process. In many instances, these technologists see only trend charts on computer screens. They may be unaware of the origin of the raw data for which they make tremendous decisions about the control of the process.
But using these methods, one may not be able to say if the process is capable of manufacturing products. Further, the data is only as good as it’s origin. Finally, what quality of the data is most important?
“Information turns”
The Most Common Mistakes Made in Parametric Test

14. Resolution and the “bins” of the normal distribution
Low Resolution rejects good die
High Resolution passes good die
Vt to
.01 V
Vt to
.005 V
So we can deduce that repeatability is more important and realistic than accuracy for the control of the semiconductor process. Repeatability is easily quantified as the standard deviation of the measured distribution.
But what about the concept of resolution. How important is that? Once the process parameter is routinely sampled the distribution of measurements can be analyzed using statistical processes. The result is that some products will be rejected for failing the test while others pass and are sold to your customer.
Those products that fail can usually be found at the extreme edges of the distribution of measured values for the process control parameter. It is in this region that products are identified as failing and therefore are rejected. The differences between high and low resolution are best exposed at the edges of the normal distribution. A finer resolution could mean fewer parts are disqualified in error!
Resolution defines the size of the window, or bin limit between good and failing die. Care must be taken to obtain the smallest possible resolution to assure passing as many good die as possible.
The Most Common Mistakes Made in Parametric Test

15. The Nature of Repeatability
Low Accuracy 2 1 3 4
High Accuracy
This slide depicts “Juran’s Target Analogy”, describing the difference between accuracy and repeatability. The analogy is of an archer shooting arrows at a target. One archer can’t hit the center of the target at all, another can hit it sometimes, yet another hits the same spot repeatedly but never hits the target, while the expert can hit the center of the target every time.
Accuracy, defined as “the extent to which the average of the measurements deviates from the true value”, requires a standard to compare against. Such standards may be obtained from theory or from standards organizations such as VLSI Standards or the National Institute for Standards and Technology (NIST).
Precision or repeatability, defined as “the measure of natural variation of repeated measurements”, can be estimated with the standard deviation of the measurement distribution. Repeatability can also be defined as “the total variation in the measurement system”. This definition includes process, spatial, operational, and measurement sources of variance. Repeatability also implies another definition. Bias, “the distance between the average value of all measurements and the true value”, becomes very important when the data is analyzed over time or when no standards are available..
So, which quadrant of the following is optimally best: (1), (2), (3), or (4)? Why?
_______________________________________________________________
From SEMATECH’s “”Introduction to Measurement Capability Analysis, 1991”.
Low Repeatability
High Repeatability
The Most Common Mistakes Made in Parametric Test

16. Lord Kelvin had something to say about statistics and measurement...
“I often say that when you can measure what you are speaking about, and express it in numbers, you know something about it …”
“But when you can not measure it, when you can not express it in numbers, your knowledge is of a meager and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to the stage of science, whatever the matter may be.”
William Thomson, Lord Kelvin, 1883
William Thomson, Lord Kelvin ( )
The first D.C. Parametric Test Engineer?
“Born in Belfast and raised in Glasgow, Thomson attended Cambridge University but returned to Scotland when he was appointed to Glasgow University's chair of natural philosophy at the tender age of twenty-two. Widely known as a mathematician, physicist and inventor, he was ennobled for his services to the British Empire in the field of navigation and the laying of the Atlantic telegraph cable. Following the spread of electric power in the 1880s, his numerous accurate and convenient measuring instruments were marketed worldwide through an elaborate system of patents and partnerships.”
Picture and text courtesy of The Archives, California Institute of Technology.
“Greatly interested in the improvement of physical instrumentation, he designed and implemented many new devices, including the mirror-galvanometer that was used in the first successful sustained telegraph transmissions in transatlantic submarine cable. He was knighted in 1866 for his work on the transatlantic cable. Thomson published more than 600 papers, was elected to the Royal Society in 1851 and served as its president from to 1895.”
From MATHEMATICAL INSTITUTE of the TECHNICAL UNIVERSITY OF BUDAPEST
Other quotes by Royal Society president Lord Kelvin,
"Radio has no future.” "X-rays are clearly a hoax.”
"The aeroplane is scientifically impossible.”
From Cambridge Career Services, Inc.
The Most Common Mistakes Made in Parametric Test

17. Top Six Reasons for Using DC Parametric Test on Electrical Test Structures
Facilitate a ‘Kelvinist approach’ to processing by expressing results in numbers thus leading to clear decision making
Provide a ‘divide and conquer approach’ to semiconductor fabrication by allowing the measurement of fundamental quantities
Allow ‘Accelerated Testing’ and measurement of parameters for reliability models
Provide a ‘common link’ between fabs to judge and improve process quality
Key Element in the rapid development of new process and new IC
Key element when transferring fabs to a new location
By Martin G. Buehler
The Most Common Mistakes Made in Parametric Test

18. Parametric Test is used for…
In Summary DC
Parametric Test is used for…
For fast development of new manufacturing processes and devices
For fast release of new processes and devices to production lines
For fast yield enhancement and process optimization to minimize
process cost
For achieving higher process technology and device reliability
And semiconductor process control requires statistical:
accuracy
repeatability
resolution
We must be careful when making measurements, we can’t make mistakes!
The Most Common Mistakes Made in Parametric Test

19. The Most Common Mistakes Made in Parametric Test*
Section 2
The Most Common Mistakes Made in Parametric Test*
*And How to Avoid Them
The Most Common Mistakes Made in Parametric Test

20. The Top Ten List Improper triaxial to coaxial adapters
Connecting SMUs up using the ‘Sense’ line instead of the ‘Force’ line
Improper connection on ground unit (GNDU)
Using ‘Limited Auto’ ranging instead of ‘Auto’ ranging
Using Default (Maximum) Current Compliance When Making Measurements in ‘Auto’ or ‘Limited Auto’ ranging
Not using SMU pulse mode, fixed measurement range, and/or Kelvin mode for high-power measurements
Using ‘Auto’ ranging in Time Sampling Mode (4155/4156)
Improperly connecting SMUs in parallel
Failure to account for electro-motive force (EMF) on sensitive voltage measurements
Improper Capacitance Measurement Techniques When Using a Switching Matrix
The Most Common Mistakes Made in Parametric Test

21. Improper triaxial to coaxial adapters
Mistake #1:
Improper triaxial to coaxial adapters
The Most Common Mistakes Made in Parametric Test

22. Why Use Triax Cables? Required For Measurements < 1nA.
Shield (Ground)
Guard
Force
Shield (Ground)
Force
Vg Triax Cables
MOSFET Subthreshold Id fA pA nA uA mA
Leakage
Vg Coax Cables
MOSFET Subthreshold Id fA pA nA uA mA
The triax cable is a special low dielectric loss, high impedance cable. This cable may be used down to fA levels when properly used with a guarded probe. The guard voltage tracks the force voltage exactly, so that no voltage drop can exist between guard and force. This eliminates the capacitive loading that would otherwise limit low current measurements.
If low impedance coax cables are used with outer layer at ground potential, two limitations will be immediately apparent. The cable leakage will limit the low current measurement floor. In addition, when the voltage is swept, the sudden change will cause additional cable charging. This distorts the low current portion of a MOS Subthreshold curve as shown.
RULES:
Unguarded coax cable is OK for measurements above 1nA.
Triax cable or coax with outer layer at guard potential should be used for measurements below 1 nA.
Eliminate cable leakage and charging currents.
The Most Common Mistakes Made in Parametric Test

23. Triaxial Guard Connection Simplified Diagram
The guard voltage tracks the force voltage exactly.
Cable charging current and noise is eliminated.
Do not ever short the guard to the force line or shield line.
Buffer x1 Rs
10K
Guard
Force
SMU
DUT
The guard connection is needed for measurement < 1 nA. Below 1 nA a regular coax cable's capacitance dominates over the DUT (device under test) capacitance. What you see is cable charging current. I = C(dv/dt) where dv/dt is the rate of change of SMU voltage from one step to the next of a coax cable with no guard. C is the total capacitance of the cable.
The above diagram shows how the cable capacitance is eliminated with a triaxial cable. The guard is driven at the same voltage as the force center conductor. No current can flow between guard and force when they are held at the same potential. Guard and force are isolated by a buffer amplifier. They can never be shorted together. V
The Most Common Mistakes Made in Parametric Test

24. How Do I Connect Triaxial and Coaxial Connections?
???
Force / Sense
Line
Driven Guard
Ground Shield
Ground Shield
Signal
What do I do with the driven guard?
Does the current I am measuring affect how I connect to a BNC connector?
Where can I get the necessary TRIAX to BNC connectors?
The Most Common Mistakes Made in Parametric Test

25. Triaxial to Coaxial Adapters: Measuring Currents > 1 nano-Amp
In this case it is OK to float the guard connection, since current leakage between the center conductor and the outer ground shield does not significantly impact the measurement.
The Most Common Mistakes Made in Parametric Test

26. Triaxial to Coaxial Adapters: Measuring Currents < 1 nano-Amp
The only way to maintain low-current measurement accuracy in a coaxial environment is to connect the driven guard to the outer shield of the coaxial connector. This presents a potential safety hazard and must be done with great care.
Warning! Shock Hazard!
The Most Common Mistakes Made in Parametric Test

27. Summary of Agilent Connectors
The Most Common Mistakes Made in Parametric Test

28. Connecting SMUs up using the ‘Sense’ line instead of the ‘Force’ line
Mistake #2:
Connecting SMUs up using the ‘Sense’ line instead of the ‘Force’ line
The Most Common Mistakes Made in Parametric Test

29. Most SMUs Have Both Force & Sense Outputs – Which Do I Use?
SMU Force Output
SMU Sense Output
If making Kelvin measurements, use both the Force and Sense outputs
If not making Kelvin measurements, always use the Force output (never use the Sense output by itself)
The Most Common Mistakes Made in Parametric Test

30. Guard and Kelvin Connection Simplified Diagram
The sense line is added.
Cable resistance error is eliminated. Useful if the DUT <50 Ohms.
Buffer x1
Guard
Sense Rs
10K
Force
SMU
DUT
The only difference between the 4155 and 4156 cable configuration is the addition of the sense line. In this case, sensing is done at the DUT, eliminating the fraction of an ohm of cable resistance. The internal sensing resistor Rs is the only feedback path in the 4155.
Note that the 4156 operates just fine without the sense cable. Then it operates just like the This is important to know because in general you do not need the sensing Kelvin connection. Most MOS measurements are high impedance and the residual cable loss is insignificant. V
The Most Common Mistakes Made in Parametric Test

31. Do Not Use SMU Sense Output by Itself!
Source / Monitor Unit (SMU):
SMU Force Output
SMU Force Circuitry
Iout
~10 KW
SMU Sense Circuitry
High Impedance
SMU Sense Output
The Most Common Mistakes Made in Parametric Test

32. Nifty Trick: Use the Sense Line as a High Impedance Scope Probe!x1
Buffer
SMU
Guard
Sense
Force V
Source
Gate
Drain
Substrate
Measure Gate Voltage versus Time Accurately
Triax to Coax Adapter
(guard floating) To
Scope
The sense line need not be used only for Kelvin connections.
It is ideal for monitoring the voltage on your device with an oscilloscope.
The sense line tracks the force line within 1mv.
All you need is a floating guard coax adapter attached to the sense line at the back of the Then use any BNC cable to direct connect the SMU sense line to the oscilloscope input.
The adapter shown is the Trompeter Electronics AD-BJ20-E2-PL75.
The Most Common Mistakes Made in Parametric Test

33. Improper connection on ground unit (GNDU)
Mistake #3:
Improper connection on ground unit (GNDU)
The Most Common Mistakes Made in Parametric Test

34. What is the Ground Unit (GNDU) Configuration?
Standard Triaxial Connection:
Ground Unit Connection:
Force / Sense
Line
Driven Guard
Ground Shield
Sense Line
Force Line
Ground Shield
The Most Common Mistakes Made in Parametric Test

35. Why is the GNDU Configuration the Way It Is?
???
Shield (Ground)
Force
Sense
In standard triaxial connections the middle conductor is a driven guard, which eliminates any cable leakage current by always keeping the driven guard the same potential as the center Force/Sense line.
In the case of the ground unit the potential of the Force and Sense lines is always at zero volts, so there is no need to shield it from the outer ground shield to prevent leakage currents.
The Most Common Mistakes Made in Parametric Test

36. What Happens if I Connect the GNDU to a Standard Triaxial Connection?
Isink
GNDU Force Circuitry
GNDU Sense Circuitry
~10 KW
High Impedance
Ground Shield
Force Line
Driven Guard
Force
Sense
Connecting a standard triaxial connector to the GNDU without an adapter is equivalent to connecting up to the SMU Sense output !
The Most Common Mistakes Made in Parametric Test

37. Proper GNDU Connection
Unless your equipment is designed to handle the GNDU connection, you must use an adapter that splits out the GNDU Force and Sense lines into standard triaxial configurations.
The Agilent N1254A-100 Ground Unit to Kelvin Adapter will split the Force and Sense lines into the proper Kelvin configuration.
GNDU
Sense Output
Force Output
The Most Common Mistakes Made in Parametric Test

38. Connections to the GNDU Should be Kelvin
Remember! Pumping large currents through cables will cause an Ohmic drop unless this is compensated via a Kelvin measurement configuration. Since assumedly the reason you are using the GNDU is to sink large currents, you should always connect up both the Force and Sense lines.
Isense (0 Amps)
GNDU
Sense Output
Force Output
Isink (Up to 4 Amps*)
*E5270A
The Most Common Mistakes Made in Parametric Test

39. Using ‘Limited Auto’ ranging instead of ‘Auto’ ranging
Mistake #4:
Using ‘Limited Auto’ ranging instead of ‘Auto’ ranging
The Most Common Mistakes Made in Parametric Test

40. Ranging – What is It?
Limited Ranging – Never go below the specified range limit
Auto Ranging – Go as low as necessary to make an accurate measurement (down to the lowest range supported if necessary)
100 mA
10 mA
1 nA
Fixed Ranging – Always stay in the same measurement range
100 pA
10 pA
The Most Common Mistakes Made in Parametric Test

41. 4156 Example
The 4156 has two additional low-current measurement ranges not available on the 4155: 100 pA & 10 pA
The 4156 boots-up like a 4155C (all SMUs set to Limited 1 nA ranging)
Unless these are changed, you cannot get the full low-current measurement capability of the instrument!
The Most Common Mistakes Made in Parametric Test

42. ID-VG Low-level Subthreshold Measure Setup Page - Default
These are 4156 default settings at “boot-up”
Measurements are made quickly but noise level is high.
Notice the LIMITED 1nA settings. These are 4155 defaults.
The Most Common Mistakes Made in Parametric Test

43. ID-VG Low-level Subthreshold Probes Up - Default Range Setting
Noise level is high. Current measurement is limited to the 1nA range.
The Most Common Mistakes Made in Parametric Test

44. ID-VG Low-level Subthreshold Measure Setup Page - AUTO range
Set the drain SMU range setting to AUTO.
This uses the two lower ranges of the 4156 and decreases the noise floor by 100x.
The Most Common Mistakes Made in Parametric Test

45. ID-VG Low-level Subthreshold Probes Up - ZERO Check
Integration time can be short.
Wait 30 minutes after boot up of 4156.
If the trace is not centered at 0 fA, press the “GREEN” key and “Zero” key to remove offset error.
+/- 2fA variation is the typical noise floor of the instrument by itself (no cables attached).
The Most Common Mistakes Made in Parametric Test

46. Extend Resolution 100X 4156C Display Page
1 fA resolution
0.01 fA resolution
This feature is only available in the 4156C. You can upgrade a 4156B to a 4156C.
The Most Common Mistakes Made in Parametric Test

47. ID-VG Low-level Subthreshold Curve Probes Down - 4156C Graph Page
Low leakage characteristics of a 40 Angstrom thick oxide n-channel MOSFET.
Resolution is .01 fA ( Amps)
The Most Common Mistakes Made in Parametric Test

48. Mistake #5:
Using Default (Maximum) Current Compliance When Making Measurements in
‘Auto’ or ‘Limited Auto’ ranging
The Most Common Mistakes Made in Parametric Test

49. Where Does the Range Search Start?
For Auto Ranging and Limited Ranging, the range search starts at the value you set for compliance.
Therefore, if you are measuring a small current and you do not change the default compliance setting (100 mA in this example), then the instrument will take longer to reach its final measurement range.
100 mA
10 nA
1 nA
100 pA
10 pA
The Most Common Mistakes Made in Parametric Test

50. How Do I Change This?
100 mA
If you know that you are measuring a small current and you want to use Auto ranging, then reduce the value of the compliance to a smaller value to speed up the measurement.
In this case, compliance was reduced to 10 nA (since a current level below 1 pA was expected).
10 nA
1 nA
100 pA
10 pA
The Most Common Mistakes Made in Parametric Test

51. Changing the 4155/4156 Compliance Settings
Reduce the compliance settings to speed up your low-current measurements
The Most Common Mistakes Made in Parametric Test

52. Remember! Need AUTO Ranging for Low Current
Keep in mind that reducing the compliance speeds up your measurement, but you still need to be using AUTO ranging in order to measure low currents (fA level)
The Most Common Mistakes Made in Parametric Test

53. Mistake #6:
Not using SMU pulse mode, fixed measurement range, and/or Kelvin mode for high-power measurements
The Most Common Mistakes Made in Parametric Test

54. SMU Pulsed Mode Reduces Thermal Heating Error
The Most Common Mistakes Made in Parametric Test

55. Kelvin Sensing Eliminates High Power Measurement Error
The Most Common Mistakes Made in Parametric Test

56. Using SMUs in Pulsed Mode Channel Definition Page
Press the “VPULSE” softkey to define the voltage pulse mode. In this case the gate SMU will be pulsed.
Note: Only ONE SMU can be in pulsed mode.
The Most Common Mistakes Made in Parametric Test

57. SMU Pulsed Mode Example of Pulsing The Primary Source
VAR Primary Source
VAR Secondary Source
Only one SMU can be pulsed.
Minimum width is 0.5 ms (includes the time to make a 5-digit resolution measurement).
The Most Common Mistakes Made in Parametric Test

58. Using SMUs in Pulsed Mode Power MOSFET Measure Page
A SMU PULSE menu appears.
Here the duty cycle of the pulse is set at 10%.
Change the PERIOD to 100ms to reduce heating further. The device will be powered on only 1% of the time.
The Most Common Mistakes Made in Parametric Test

59. Using SMUs in Pulsed Mode Measure Setup Page
VERY IMPORTANT!
Use FIXED ranges.
FIXED prevents the SMUs from auto ranging. AUTO over-rides the pulse settings and could add up to 30ms on each range change.
Here Id is set to the 1Amp range to make use of the HPSMU in the expander box.
The Most Common Mistakes Made in Parametric Test

60. Using SMUs in Pulsed Mode Graph Page
VERY IMPORTANT!
Use SHORT integration.
This minimizes heating and assures proper timing.
LONG integration will over-ride your pulse width/length setting. Much more time is required for long integration.
The Most Common Mistakes Made in Parametric Test

61. Kelvin Triax Cable Ideal for both low current and low impedance applications.
Guard
Force
Ground
Sense + -
IForce
SMU1 (Force)
SMU2 (Force)
SMU1 (Sense)
SMU2 (Sense)
VSense
I=0
The 4155 uses the same triax cables as the 4142 and These cables are good for low current measurements. However, two cables are necessary for low resistance Kelvin measurements.
Agilent Technologies designed a special Kelvin triax cable for the This cable is optimized for both low current and low resistance measurement. Both force and sense lines are held rigidly in the same Teflon cable. Friction is reduced and the cable is less sensitive to noise caused by moving the cable.
Kelvin triax cable assemblies are available with two connector options:
16434A compatible on one end; compatible on the other end
16493K 4156 compatible on both ends (standard option)
The Most Common Mistakes Made in Parametric Test

62. Non-Kelvin Measurements Can Introduce Significant Error
Slope = 1/Re
(Kelvin)
Slope = 1/(Re+Rcable)
(Non-Kelvin) I B
Re = 0.55
Rcable = 0.40 V C
monitor
Cable resistance comparable to resistance being measured
In the example above, the device is connected with a SMU on the base sweeping current, a voltmeter on the collector, and the emitter is grounded with a Kelvin SMU. The base SMU does not have to be Kelvin since we are only forcing current and do not care about measuring the cable loss in the base. Also, the collector SMU is being used only as a high impedance voltmeter, so there is no cable loss in this lead.
The emitter on the other hand, must be connected to a Kelvin SMU. Because of this, we can compensate for the 0.40 ohm path through the cable and fixture. From the graph we can see the emitter resistance is 0.55 ohm when compensated using the Kelvin connection. Non-Kelvin resistance is 0.95 ohm, due to the extra 0.40 ohm cable and fixture resistance error.
The Most Common Mistakes Made in Parametric Test

63. Wafer Prober Kelvin Cable Connections Optimized For Measurement Accuracy
To Kelvin Probe
Photo of SMU cable connection to a Cascade Microtech Summit probe station.
4156's Kelvin triaxial cables mate directly to up to six probes top side and a guarded Kelvin chuck (substrate) connection. There is even a provision for mating the 1.6 A ground unit (GNDU) of the expander box.
This station uses the Micro Chamber (TM) design for a small volume shielded box enclosing only the probes and wafer; not the entire probe station. The rigid mechanical design with guarded chuck provides an ideal environment for fA current, fF capacitance, and uV voltage measurements.
To Guarded Chuck
The Most Common Mistakes Made in Parametric Test

64. Using ‘Auto’ ranging in Time Sampling Mode (4155/4156)
Mistake #7:
Using ‘Auto’ ranging in Time Sampling Mode (4155/4156)
The Most Common Mistakes Made in Parametric Test

65. What is Time Sampling?
60 ms to 65 sec
(4155/4156)
Initial Interval
Time
Time sampling involves measuring a voltage or current at regular intervals over time.
Useful for certain types of reliability stress measurements such as Time Dependent Dielectric Breakdown (TDDB)
The Most Common Mistakes Made in Parametric Test

66. Specifying a Small Sampling Interval and Auto Ranging Creates a Conflict!
Time sampling inherently requires that the measurement occur within a certain time period. Otherwise, the sampling rate cannot be met.
Auto ranging requires the instrument to start at the specified compliance value and work its way down to the correct measurement range, which takes time.
Specifying both a short sampling time (fast sample rate) and auto ranging creates a conflict for the instrument!
The Most Common Mistakes Made in Parametric Test

67. How Does this Conflict Get Resolved?
Accuracy always wins out over speed
The instrument will take as long as necessary to auto range, ignoring the specified measurement interval settings
The end result is that the instrument will not measure at the interval you specified!
NEVER USE AUTO-RANGING WHEN MAKING FAST TIME SAMPLING MEASUREMENTS
The Most Common Mistakes Made in Parametric Test

68. How Do I Optimize My Time Sampling Measurements?
???
Besides using FIXED ranging, what else do I need to do to optimize my Time Sampling measurements?
Minimize active units
Measure on only one resource
Disable the STOP condition
Minimize the compliance setting
The Most Common Mistakes Made in Parametric Test

69. Optimizing Time Sampling Measurements - 1
Minimize the number of active resources to only those that you need.
The Most Common Mistakes Made in Parametric Test

70. Optimizing Time Sampling Measurements - 2
For intervals < 2 ms, must have STOP CONDITION set to DISABLE
For intervals < 2 ms, can only have one measurement channel
Make sure that compliance is set as low as possible
The Most Common Mistakes Made in Parametric Test

71. Optimizing Time Sampling Measurements - 3
For intervals < 2 ms, must used FIXED measurement range (never use AUTO ranging).
The Most Common Mistakes Made in Parametric Test

72. Improperly connecting SMUs in parallel
Mistake #8:
Improperly connecting SMUs in parallel
The Most Common Mistakes Made in Parametric Test

73. Why Would I Connect SMUs in Parallel?
???
Connecting SMUs in parallel allows you to increase the total current delivered to a DUT
SMU 1
SMU 2
DUT
ITotal = ISMU1 + ISMU2
The Most Common Mistakes Made in Parametric Test

74. I Force, V Measure (Non-Kelvin Connection)
Sense
SMU 1 Vm
SMU 2
DUT
Easy to do: Can control with I/CV
Voltage measurement accuracy is relatively poor
The Most Common Mistakes Made in Parametric Test

75. I Force, V Measure (Kelvin Connection)
Sense
SMU 1 Vm
SMU 2
DUT
Easy to do: Can control with I/CV
High voltage measurement accuracy
The Most Common Mistakes Made in Parametric Test

76. Current Force Mode is not Always Useful
Sense
SMU Vm
Paralleling SMUs in current source mode is easy, BUT:
Not all applications can be covered this way
It begs the question of how to parallel SMUs in voltage force mode
The Most Common Mistakes Made in Parametric Test

77. V Force, I Measure (Non-Kelvin Connection)
Sense
SMU 1
SMU 2
DUT
Voltage force accuracy is poor
Easy to use in concept, BUT requires great care to prevent two voltage sources from interfering with one another
The Most Common Mistakes Made in Parametric Test

78. V Force, I Measure (Kelvin Connection)
Sense
SMU 1 Vm
SMU 2
DUT
High voltage force accuracy
Requires sophisticated measurement control software
The Most Common Mistakes Made in Parametric Test

79. Mistake #9:
Failure to account for electro-motive force (EMF) on sensitive voltage measurements
The Most Common Mistakes Made in Parametric Test

80. Thermo Electro-Motive Force (EMF): What is It?
A transient voltage pulse that is associated with reed relay switches.
Note: This is NOT an example of the relays used in our instrumentation.
Conventional reed relay switches, which can be obtained from a variety of sources, typically generate a thermo-EMF (electro-motive-force) ranging from a few tens of micro-volts to a few hundreds of micro-volts after the relay activation current is turned on or off. This voltage drift, which can continue for several minutes before dying out, is usually not acceptable when making precision measurements such as those required for BJT matching characterization. The above figure shows an example of the thermal-EMF generated by a commercially available reed relay.
The Most Common Mistakes Made in Parametric Test

81. Agilent 4073A/B Relays Eliminate Thermo-EMF
Agilent 4073A example through the switching matrix
The Agilent 4073A and 4073B test systems use a proprietary reed relay that almost completely eliminates the thermo-EMF problem. The above graph illustrates the dramatic difference between the performance of the 4073A/B relays versus those shown on the previous slide. As you can see, the relays in the 4073A/B act as near ideal switches.
The Most Common Mistakes Made in Parametric Test

82. Managing the Use of a Reed Relay Switch
Wait until the thermo-EMF has stabilized to its final value.
For the 4156C and E5270, set the SMU output relay to its on (closed) state after warm-up and keep it there.
Complete the measurement as quickly as possible.
Complete the measurement within 10 seconds. This keeps the total drift to less than a few micro-volts.
The reed relays used in semiconductor parameter analyzers and switching matrices are not as close to the ideal case as are those used in the 4073A/B. The data sheet specifications of the parameter analyzer SMUs take the thermo-EMF effects into account so users do not have to worry about this effect for normal applications. However, when performing measurement for matching applications that require extremely high levels of accuracy beyond the normal specifications, the guidelines shown above can minimize or eliminate the thermo-EMF effects.
Note that the fine control necessary to manage the reed relay output switch on the 4156C can only be accomplished using the FLEX programming mode. If you are controlling the 4156C with ICS or I/CV, then you should create a VBScript algorithm utilizing the FLEX programming mode to make sure that the SMU output relays remain closed and stable throughout your matching measurement.
The Most Common Mistakes Made in Parametric Test

83. Kelvin Resistance Measurement
Key Points:
Make sure that you eliminate Joule self-heating effects
Measure and subtract Voffset
Measure twice with opposing current flow and average the two resistances
R = (VM1 – VM2)/IF
VMU 2
(SMU 4)
SMU 2
VMU 1
(SMU 3)
SMU 1 R
Force Current Twice
(Both Ways), i.e. +/- IF
VM1
VM2
0 V
VMU 1
(SMU 3)
VM1
VEMF1
VOFF1 + -
VMU 2
(SMU 4)
VM2
VEMF2
VOFF2
A resistor can be measured accurately in two ways: using the Kelvin (or 4-terminal) measurement method with a precision resistance meter, or using SMUs and VMUs as shown in the above figure. To obtain a successful resistance measurement, two things are important: 1) elimination of the Joule self-heating effect, which will increase the temperature of the device, and 2) measuring twice, which requires applying current in both directions by switching the polarity of the force current (IF). By measuring twice, you can take the average of the two resistances to cancel the offset voltage of the VMU (or voltage sense) and the thermo-EMF of the connection terminal. The easiest way to measure resistors is by using the SMU/VMU method, because the current (effectively the power) applied to the resistor can be controlled.
Note: It is left as an exercise for the reader to convince yourself that by applying Kirchoff’s current law and voltage law to the above circuit for the two different cases involving the force current and then averaging the two calculated R values, the effects of the thermo-EMF and voltage offset are eliminated.
The Most Common Mistakes Made in Parametric Test

84. Mistake #10:
Improper Capacitance Measurement Techniques When Using a Switching Matrix
The Most Common Mistakes Made in Parametric Test

85. Three Most Common Mistakes Using a Four terminal pair through a Switching Matrix:
2. Ignoring return path connection Hc Hp V Lc A Lp
MATRIX
3. Cable impedance not 50 Ohms.
1. Unsupported cable length (> 4 meters)
The Most Common Mistakes Made in Parametric Test

86. Mistake 1: Unsupported Cable Length Causes Error
The 4284A can only compensate cable lengths up to 4m
The innate 4284A compensation routine only supports cable lengths up to 4 meters. Additional correction is needed for longer cable lengths. Hc Hp V
100pF/m 250nH/m Lc A
100pF/m 250nH/m Lp
MATRIX
If the cable length is too long, then the 4284A may not be able to balance its bridge!
The Most Common Mistakes Made in Parametric Test

87. Open/Short/Load Correction Model
Ideal
Assume I1=i1’
C meter I1 I2
CMH V1 V2
DUT
CML
I1’
The built-in 4284A compensation algorithm extracts four parameters using three (open/short/load) measurements.
If the cable length is too long…
I1~i1’
C meter I1 I2
CMH
Guard V2
DUT
CML
Guard
I1’
Cables that are too long invalidate the assumptions. The number of port parameters becomes greater than four and the built-in 4284A compensation routine does not work.
The Most Common Mistakes Made in Parametric Test

88. Limitation of OPEN/SHORT/LOAD correction
Typical parametric measurements are in this range.
Open/Short/Load correction is not perfect. Load impedance should be same range as the device impedance you want to measure.
The Most Common Mistakes Made in Parametric Test

89. Mistake 2: No Return Path Wire Causes Error
Return wire stabilizes the cable inductance.
Make return path near the device. Hc Hp V Lc A Lp
Return Wire length is also important.
MATRIX
Cable inductance can change from 250 nH/m to more than 400 nH/m by removing return wire.
The Most Common Mistakes Made in Parametric Test

90. Mistake 3: Wrong Cable Causes Measurement Error
4284A calibration routine expects 50 Ohm cable environment
333 nH/m 250nH/m Hc
The built-in 4284A correction routine supports only 50 Ohm cables. Need additional correction for cables that are not 50 Ohms. Hp V Lc A Lp
MATRIX
Agilent Triaxial cable is not 50 Ohm
Matrix internal path is not 50ohm
The Most Common Mistakes Made in Parametric Test

91. Conclusions / Summary
A little understanding can go a long way towards helping you improve your parametric measurement skills.
More information is available in our “Parametric Test Assistant CD” (Agilent Publication # EN)
Live phone-based assistance is available by calling Agilent’s US Customer Care Center at:
The Most Common Mistakes Made in Parametric Test