With the increasing clock speeds and the decreasing feature sizes found in today's nanometer designs, at-speed testing is a requirement to achieve high quality test results. In addition, new advanced fault models are also available to improve defect detection and lower DPM rates. Advanced at-speed test capabilities and some new fault models are described in this paper.
Showing posts with label DFT. Show all posts
Showing posts with label DFT. Show all posts
The Economics of Test, Part - IV
Detecting a defective unit is often only part of the job. Another important aspect of test economics that must be considered is the cost of locating and replacing defective parts. Consider again the board with 10 integrated circuits. If it is found to be defective, then it is necessary to locate the part that has failed, a time-consuming and error-prone operation. Replacing suspect components that have been soldered onto a PCB can introduce new defects. Each replaced component must be followed by retest to ensure that the component replaced was the actual failing component and that no new defects were introduced during this phase of the operation. This ties up both technician and expensive test equipment. Consequently, a goal of test development must be to create tests capable of not only detecting a faulty operation but to pinpoint, whenever possible, the faulty component. In actual practice, there is often a list of suspected components and the objective must be to shorten, as much as possible, that list.
One solution to the problem of locating faults during the manufacturing process is to detect faulty devices as early as possible. This strategy is an acknowledgment of the so-called rule-of-ten. This rule, or guideline, asserts that the cost of locating a defect increases by an order of magnitude at every level of integration. For example, if it cost N dollars to detect a faulty chip at incoming inspection, it may cost 10N dollars to detect a defective component after it has been soldered onto a PCB. If the component is not detected at board test, it may cost 100 times as much if the board with the faulty component is placed into a complete system. If the defective system is shipped to a customer and requires that a field engineer make a trip to a customer site, the cost increases by another power of 10. The obvious implication is that there is tremendous economic incentive to find defects as early as possible. This preoccupation with finding defects early in the manufacturing process also holds for ICs.27 A wafer will normally contain test circuits in the scribe lanes between adjacent die. Parametric tests are performed on these test circuits. If these tests fail, the wafer is discarded, since these circuits are far less dense than the circuits on the die themselves. The next step is to perform a probe test on individual die before they are cut from the wafer. This is a gross test, but it detects many of the defective die. Those that fail are discarded. After the die are cut from the wafer and packaged, they are tested again with a more thorough functional test. The objective? Avoid further processing, and subsequent packaging, of die that are clearly defective.
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
One solution to the problem of locating faults during the manufacturing process is to detect faulty devices as early as possible. This strategy is an acknowledgment of the so-called rule-of-ten. This rule, or guideline, asserts that the cost of locating a defect increases by an order of magnitude at every level of integration. For example, if it cost N dollars to detect a faulty chip at incoming inspection, it may cost 10N dollars to detect a defective component after it has been soldered onto a PCB. If the component is not detected at board test, it may cost 100 times as much if the board with the faulty component is placed into a complete system. If the defective system is shipped to a customer and requires that a field engineer make a trip to a customer site, the cost increases by another power of 10. The obvious implication is that there is tremendous economic incentive to find defects as early as possible. This preoccupation with finding defects early in the manufacturing process also holds for ICs.27 A wafer will normally contain test circuits in the scribe lanes between adjacent die. Parametric tests are performed on these test circuits. If these tests fail, the wafer is discarded, since these circuits are far less dense than the circuits on the die themselves. The next step is to perform a probe test on individual die before they are cut from the wafer. This is a gross test, but it detects many of the defective die. Those that fail are discarded. After the die are cut from the wafer and packaged, they are tested again with a more thorough functional test. The objective? Avoid further processing, and subsequent packaging, of die that are clearly defective.
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
The Economics of Test, Part - III
However, if devices are tested, feature sizes can be reduced and more die will fit on each wafer. Even after the die are tested and defective die are discarded, the number of good die per wafer exceeds the number available at the larger feature sizes. The benefit in terms of increasing numbers of good die obtainable from each wafer far outweighs the cost of testing the die in order to identify those that are defective. Point B on the graph corresponds to a point where process yield is lower than the required quality level. However, testing will identify enough defective units to bring quality back to the required quality level. The horizontal distance from point A to point B on the graph is an indication of the extent to which the process capability can be made more aggressive, while meeting quality goals. The object is to move as far to the right as possible, while remaining competitive. At some point the cost of test will be so great, and the yield of good die so low, that it is not economically feasible to operate to the right of that point on the solid line.
We see therefore that we are caught in a dilemma: Testing adds cost to a product, but failure to test also adds cost. Trade-offs must be carefully examined in order to determine the right amount of testing. The right amount is that amount which minimizes total cost of testing plus cost of servicing or replacing defective components. In other words, we want to reach the point where the cost of additional testing exceeds the benefits derived. Exceptions exist, of course, where public safety or national security interests are involved.
Another useful side effect of testing that should be kept in mind is the information derived from the testing process. This information, if diligently recorded and analyzed, can be used to learn more about failure mechanisms. The kinds of defects and the frequency of occurrence of various defects can be recorded and this information can be used to improve the manufacturing process, focusing attention on those areas where frequency of occurrence of defects is greatest.
This test versus cost dilemma is further complicated by “time to market.” Quality is sometimes seen as one leg of a triangle, of which the other two are “time to market” and “product cost.” These are sometimes posited as competing goals, with the suggestion that any two of them are attainable.25 The implication is that quality, while highly desirable, must be kept in perspective. Business Week magazine, in a feature article that examined the issue of quality at length, expressed the concern that quality could become an end in itself. The importance of achieving a low defect level in digital components can be appreciated from just a cursory look at a typical PCB. Suppose, for example, that a PCB is populated with 10 components, and each component has a defect level DL = 0.999. The likelihood of getting a defect free board is (0.999)10 = 0.99004; that is, one of every 100 PCBs will be defective—and that assumes no defects were introduced during the manufacturing process. If several PCBs of comparable quality go into a more complex system, the probability that the system will function correctly goes down even further.
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
We see therefore that we are caught in a dilemma: Testing adds cost to a product, but failure to test also adds cost. Trade-offs must be carefully examined in order to determine the right amount of testing. The right amount is that amount which minimizes total cost of testing plus cost of servicing or replacing defective components. In other words, we want to reach the point where the cost of additional testing exceeds the benefits derived. Exceptions exist, of course, where public safety or national security interests are involved.
Another useful side effect of testing that should be kept in mind is the information derived from the testing process. This information, if diligently recorded and analyzed, can be used to learn more about failure mechanisms. The kinds of defects and the frequency of occurrence of various defects can be recorded and this information can be used to improve the manufacturing process, focusing attention on those areas where frequency of occurrence of defects is greatest.
This test versus cost dilemma is further complicated by “time to market.” Quality is sometimes seen as one leg of a triangle, of which the other two are “time to market” and “product cost.” These are sometimes posited as competing goals, with the suggestion that any two of them are attainable.25 The implication is that quality, while highly desirable, must be kept in perspective. Business Week magazine, in a feature article that examined the issue of quality at length, expressed the concern that quality could become an end in itself. The importance of achieving a low defect level in digital components can be appreciated from just a cursory look at a typical PCB. Suppose, for example, that a PCB is populated with 10 components, and each component has a defect level DL = 0.999. The likelihood of getting a defect free board is (0.999)10 = 0.99004; that is, one of every 100 PCBs will be defective—and that assumes no defects were introduced during the manufacturing process. If several PCBs of comparable quality go into a more complex system, the probability that the system will function correctly goes down even further.
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
The Economics of Test, Part - II
The table depicted shows test cost broken down into four categories some of which are one-time, non recurring costs whereas others are recurring costs. Test preparation includes costs related to development of the test programs as well as some potential costs incurred during design of the DFT features.
DFT-related costs are directed toward improving access to the basic functionality of the design in order to simplify the creation of test programs. Many of the factors depicted in the Figure imply both recurring and nonrecurring costs. Test execution requires personnel and equipment. The tester is amortized over individual units,
representing a recurring cost for each unit tested, while costs such as probe cards may represent a one-time, nonrecurring cost. The testrelated silicon is a recurring cost, while the design effort required to incorporate testability enhancements, listed under test preparation as DFT design, is a nonrecurring cost.
The category listed as imperfect test quality includes a subcategory labeled as tester escapes, which are bad chips that tested good. It would be desirable for tester escapes to fall in the category of nonrecurring costs but, regrettably, tester escapes are a fact of life and occur with unwelcome regularity.
Lost performance refers to losses caused by increases in die size necessary to accommodate DFT features. The increase in die size may result in fewer die on a wafer; hence a greater number of wafers must be processed to achieve a given throughput. Lost yield is the cost of discarding good die that were judged to be bad by the tester.
The column in Figure labeled “Volume” is a critical factor. For a consumer product with large production volumes, more time can be justified in developing a comprehensive test plan because development costs will be amortized over many units. Not only can a more thorough test be justified, but also a more efficient test—that is, one that reduces the amount of time spent in testing each individual unit. In low-volume products, testing becomes a disproportionately large part of total product cost and it may be impossible to justify the cost of refining a test to make it more efficient. However, in critical applications it will still be necessary to prepare test programs that are thorough in their ability to detect defects.
A question frequently raised is, “How much testing is enough?” That may seem to be a rather frivolous question since we would like to test our product so thoroughly that a customer never receives a defective product. When a product is under warranty or is covered by a service contract, it represents an expense to the manufacturer when it fails because it must be repaired or replaced. In addition, there is an immeasurable cost in the loss of customer goodwill, an intangible but very real cost, not reflected in the Figure, that results from shipping defective products. Unfortunately we are faced with the inescapable fact that testing adds cost to a
product. What is sometimes overlooked, however, is the fact that test cost is recovered by virtue of enhanced throughput. Consider the graph in the Figure. The solid line reflects quality level, in terms of defects per million (DPM) for a given process, assuming no test is performed. It is an inverse relationship; the higher the required quality, the fewer the number of die obtainable from the process. This follows from the simple fact that, for a given process, if higher quality (fewer DPM) is required, then feature sizes must be increased. The problem with this manufacturing model is that, if required quality level is too high, feature sizes may be so large that it is impossible to produce die competitively. If the process is made more aggressive, an increasing number of die will be defective, and quality levels will fall. Point A on the graph corresponds to the point where no testing is performed. Any attempt to shrink the process to get more units per wafer will cause quality to fall below the required quality level.
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
DFT-related costs are directed toward improving access to the basic functionality of the design in order to simplify the creation of test programs. Many of the factors depicted in the Figure imply both recurring and nonrecurring costs. Test execution requires personnel and equipment. The tester is amortized over individual units,
representing a recurring cost for each unit tested, while costs such as probe cards may represent a one-time, nonrecurring cost. The testrelated silicon is a recurring cost, while the design effort required to incorporate testability enhancements, listed under test preparation as DFT design, is a nonrecurring cost.The category listed as imperfect test quality includes a subcategory labeled as tester escapes, which are bad chips that tested good. It would be desirable for tester escapes to fall in the category of nonrecurring costs but, regrettably, tester escapes are a fact of life and occur with unwelcome regularity.
Lost performance refers to losses caused by increases in die size necessary to accommodate DFT features. The increase in die size may result in fewer die on a wafer; hence a greater number of wafers must be processed to achieve a given throughput. Lost yield is the cost of discarding good die that were judged to be bad by the tester.
The column in Figure labeled “Volume” is a critical factor. For a consumer product with large production volumes, more time can be justified in developing a comprehensive test plan because development costs will be amortized over many units. Not only can a more thorough test be justified, but also a more efficient test—that is, one that reduces the amount of time spent in testing each individual unit. In low-volume products, testing becomes a disproportionately large part of total product cost and it may be impossible to justify the cost of refining a test to make it more efficient. However, in critical applications it will still be necessary to prepare test programs that are thorough in their ability to detect defects.
A question frequently raised is, “How much testing is enough?” That may seem to be a rather frivolous question since we would like to test our product so thoroughly that a customer never receives a defective product. When a product is under warranty or is covered by a service contract, it represents an expense to the manufacturer when it fails because it must be repaired or replaced. In addition, there is an immeasurable cost in the loss of customer goodwill, an intangible but very real cost, not reflected in the Figure, that results from shipping defective products. Unfortunately we are faced with the inescapable fact that testing adds cost to a
product. What is sometimes overlooked, however, is the fact that test cost is recovered by virtue of enhanced throughput. Consider the graph in the Figure. The solid line reflects quality level, in terms of defects per million (DPM) for a given process, assuming no test is performed. It is an inverse relationship; the higher the required quality, the fewer the number of die obtainable from the process. This follows from the simple fact that, for a given process, if higher quality (fewer DPM) is required, then feature sizes must be increased. The problem with this manufacturing model is that, if required quality level is too high, feature sizes may be so large that it is impossible to produce die competitively. If the process is made more aggressive, an increasing number of die will be defective, and quality levels will fall. Point A on the graph corresponds to the point where no testing is performed. Any attempt to shrink the process to get more units per wafer will cause quality to fall below the required quality level.About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
The Economics of Test, Part - I
What are the factors that influence the cost of test? Quality and test costs are related, but they are not inverse of one another. As we shall see, an investment in a higher-quality test often pays dividends during the test cycle. Test related costs for ICs and PCBs include both time and resource. As pointed out in previous sections, for some products the failure to reach a market window early in the life cycle of the product can cause significant loss of revenue and may in fact be fatal to the future of the product.
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
About the Author:
Name: Joachim Bauer, Test Engineer
Experience: 13+ Yrs
Location: Nice, France
When are DFT and Formal verification used?
DFT:
- manufacturing defects like stuck at "0" or "1".
- test for set of rules followed during the initial design stage.
Formal verification:
- Verification of the operation of the design, i.e, to see if the design follows spec.
- gate netlist == RTL (Equivalence checking)
- using mathematics and statistical analysis to check for eqivalence.
Application Specific Integrated Circuit ( ASIC )
An application-specific integrated circuit or ASIC comprises an integrated circuit (IC) with functionality customized for a particular use (equipment or project), rather than serving for general-purpose use.
For example, a chip designed solely to run a cash register is an ASIC. In contrast, a microprocessor is not application-specific, because users can adapt it to many purposes.
The initial ASICs used gate-array technology.
The British firm Ferranti produced perhaps the first gate-array, the ULA (Uncommitted Logic Array), around 1980. Customisation occurred by varying the metal interconnect mask. ULAs had complexities of up to a few thousand gates. Later versions became more generalized, with different base dies customised by both metal and polysilicon layers. Some base dies include RAM elements.
In the late 1980s, the availability of logic synthesis tools (such as Design Compiler) that could accept hardware description language descriptions using Verilog and VHDL and compile a high-level description into to an optimised gate level netlist brought "standard-cell" design into the fore-front. A standard-cell library consists of pre-characterized collections of gates (such as 2 input nor, 2 input nand, invertors, etc.) that the silicon compiler uses to translate the original source into a gate level netlist. This netlist is fed into a place and route tool to create a physical layout. Routing applications then place the pre-characterized cells in a matrix fashion, and then route the connections through the matrix. The final output of the "place & route" process comprises a data-base representing the various layers and polygons in GDS-II format that represent the different mask-layers of the actual chip.
Finally, designers can also take the "full-custom" route in implementing an ASIC. In this case, an individual description of each transistor occurs in building the circuit. A "full-custom" implementation may function five times faster than a "standard-cell" implementation. The "standard-cell" implementation can usually be implemented quite a bit quicker and with less risk of errors, than the "full-custom" choice.
As feature sizes have shrunk and design tools improved over the years, the maximum complexity (and hence functionality) has increased from 5000 gates to 20 million or more. Modern ASICs often include 32-bit processors and other large building-blocks. Many people refer to such an ASIC as a SoC - System on a Chip.
The use of intellectual property (IP) in ASICs has become a growing trend. Many ASIC houses have had standard cell libraries for years. However IP takes the reuse of designs to a new level. Designers of most complex digital ICs now utilise computer languages that describe electronics rather than code. Many organizations now sell tested functional blocks written in these languages. For example, one can purchase CPUs, ethernet or telephone interfaces.
For smaller designs and/or lower production volumes, ASICs have started to become a less attractive solution, as field-programmable gate arrays (FPGAs) grow larger, faster and more capable. Some SoCs consist of a microprocessor, various types of memory and a large FPGA.
So having said, this blog is dedicated to Digital Electronics, VLSI, ASICs, SOCs etc.
For example, a chip designed solely to run a cash register is an ASIC. In contrast, a microprocessor is not application-specific, because users can adapt it to many purposes.
The initial ASICs used gate-array technology.
The British firm Ferranti produced perhaps the first gate-array, the ULA (Uncommitted Logic Array), around 1980. Customisation occurred by varying the metal interconnect mask. ULAs had complexities of up to a few thousand gates. Later versions became more generalized, with different base dies customised by both metal and polysilicon layers. Some base dies include RAM elements.
In the late 1980s, the availability of logic synthesis tools (such as Design Compiler) that could accept hardware description language descriptions using Verilog and VHDL and compile a high-level description into to an optimised gate level netlist brought "standard-cell" design into the fore-front. A standard-cell library consists of pre-characterized collections of gates (such as 2 input nor, 2 input nand, invertors, etc.) that the silicon compiler uses to translate the original source into a gate level netlist. This netlist is fed into a place and route tool to create a physical layout. Routing applications then place the pre-characterized cells in a matrix fashion, and then route the connections through the matrix. The final output of the "place & route" process comprises a data-base representing the various layers and polygons in GDS-II format that represent the different mask-layers of the actual chip.
Finally, designers can also take the "full-custom" route in implementing an ASIC. In this case, an individual description of each transistor occurs in building the circuit. A "full-custom" implementation may function five times faster than a "standard-cell" implementation. The "standard-cell" implementation can usually be implemented quite a bit quicker and with less risk of errors, than the "full-custom" choice.
As feature sizes have shrunk and design tools improved over the years, the maximum complexity (and hence functionality) has increased from 5000 gates to 20 million or more. Modern ASICs often include 32-bit processors and other large building-blocks. Many people refer to such an ASIC as a SoC - System on a Chip.
The use of intellectual property (IP) in ASICs has become a growing trend. Many ASIC houses have had standard cell libraries for years. However IP takes the reuse of designs to a new level. Designers of most complex digital ICs now utilise computer languages that describe electronics rather than code. Many organizations now sell tested functional blocks written in these languages. For example, one can purchase CPUs, ethernet or telephone interfaces.
For smaller designs and/or lower production volumes, ASICs have started to become a less attractive solution, as field-programmable gate arrays (FPGAs) grow larger, faster and more capable. Some SoCs consist of a microprocessor, various types of memory and a large FPGA.
So having said, this blog is dedicated to Digital Electronics, VLSI, ASICs, SOCs etc.