Part of the  

Solid State Technology

  and   

The Confab

  Network

About  |  Contact

Posts Tagged ‘wafer processing’

Deep Learning Joins Process Control Arsenal

Friday, December 8th, 2017

thumbnail

By David Lammers

At the 2017 Advanced Process Control (APC 2017) conference, several companies presented implementations of deep learning to find transistor defects, align lithography steps, and apply predictive maintenance.

The application of neural networks to semiconductor manufacturing was a much-discussed trend at the 2017 APC meeting in Austin, starting out with a keynote speech by Howard Witham, Texas operations manager for Qorvo Inc. Witham said artificial intelligence has brought human beings to “a point in history, for our industry and the world in general, that is more revolutionary than a small, evolutionary step.”

People in the semiconductor industry “need to take what’s out there and figure out how to apply it to your own problems, to figure out where does the machine win, and where does the brain still win?” Witham said.

At Seagate Technology, a small team of engineers stitched together largely packaged or open source software running on a conventional CPU to create a convolution neural network (CNN)-based tool to find low-level device defects.

In an APC paper entitled Automated Wafer Image Review using Deep Learning, Sharath Kumar Dhamodaran, an engineer/data scientist based at Seagate’s Bloomington, Minn. facility, said wafers go through several conventional visual inspection steps to identify and classify defects coming from the core manufacturing process. The low-level defects can be identified by the human eye but are prone to misclassification due to the manual nature of the inspections.

Each node in a convolutional layer takes a linear combination of the inputs from nodes in the previous layer, and then applies a nonlinearity to generate an output and pass it to nodes in the next layer. Source: Seagate

“Some special types of low-level defects can occur anywhere in the device. While we do have visual inspections, mis-classifications are very common. By using deep learning, we can solve this issue, achieve higher levels of confidence, and lower mis-classification rates,” he said.

CNNs, Hadoop, and Apache

The deep learning system worked well but required a fairly extensive training cycle, based on a continuously evolving set of training images. The images were replicated from an image server into an Apache HBASE table on a Hadoop cluster. The HBASE table was updated every time images were added to the image server.

To improve the neural network training steps, the team artificially created zoomed-in copies of the same image to enlarge the size of the training set. This image augmentation, which came as part of a software package, was used so that the model did not see the same image twice, he said.

“Our goal was to demonstrate the power of our models, so we did no feature engineering and only minimal pre-processing,” Dhamodaran said.

A Convolution Neural Network (CNN)

Neural networks are trained with many processing layers, which is where the term deep learning comes from. The CNN’s processing layers are sensitive to different features, such as edges, color, and other attributes. This heterogeneity “is exploited to construct sophisticated architectures in which the neurons and layers are connected, and plays a primary role in determining the network’s ability to produce meaningful results,” he said.

The model was trained initially with about 7,000 images over slightly less than six hours on a conventional CPU. “If training the model had been done on a high-performance GPU, it would have taken less than a minute for several thousand images,” Dhamodaran said.

The team used commercially available software, writing code in Python and using Ubuntu, Tensorflow, Keras and other data science packages.

After the deep learning system was put into use, the rate of false negatives on incoming images was excellent. Dhamodaran said the defect classification process was much better than the manual system, with 95 percent of defects correctly classified and the remaining five percent mis-classifications. With the manual system, images were correctly classified only 60 percent of the time.

“None of the conventional machine learning models could do what deep learning could do. But deep learning has its own limitations. Since it is a neural network it is a black box. Process engineers in a manufacturing setting would like to know ‘How does this classification happen?’ That is quite challenging.”

The team created a dashboard so that when an unseen defect occurs the system can incorporate feedback from the operator, feedback which can be incorporated in the next training cycle, or used to create the training set for different processes.

The project involved fewer than six people, and took about six months to put all the pieces together. The team deployed the system on a workstation in the fab, achieving better-than-acceptable decision latency during production.

While Dhamodaran said future implementations of deep learning can be developed in a shorter time, building on what the team learned in the first implementation. He declined to detail the number of features that the initial system dealt with.

Seagate engineer Tri Nguyen, a co-author, said future work involves deploying the deep learning system to more inspection steps. “This system doesn’t do anything but image processing, and the classification is good or bad. But even with blurry images, the system can achieve a high level of confidence. It frees up time and allow operators to do some root cause analysis,” Nguyen said.

Seagate engineers Tri Nguyen and Sharath Kuman Dhamodaran developed a deep learning tool for wafer inspection that sharply reduced mis-classifications.

Python, Keras, TensorFlow

Jim Redman, president of consultancy Ergo Tech (Santa Fe, N.M.), presented deep learning work done with a semiconductor manufacturer to automate lithography alignment. Redman was unabashedly positive about the potential of neural networks for chip manufacturing applications. The movement toward deep learning, he said, “really started” from the date — 9 November 2015 – when the TensorFlow software, developed within the Google Brain group, was released under an Apache 2.0 open source license.

Other tools have further eased the development of deep learning applications, Redman added, including Keras, a high-level neural network API, written in Python and capable of running on top of TensorFlow, for enabling fast experimentation.

In just the last year or so, the application of neural networks in the chip industry has made “huge advances,” Redman said, arguing that deep learning is “a wave that is coming. It is a transformative technology that will have a transformative effect on the semiconductor industry.”

In image processing and analysis, what is difficult to do with conventional techniques often can be handled more easily by neural networks. “The beauty of neural networks is that you can take training sets and teach the model something by feeding in known data. You train the model with data points, and then you feed in unknown data.”

While Redman’s work involved lithography alignment, he said “there is no reason the same learning shouldn’t apply to etch tools or electroplaters. It is the basically the same model.”

Less Code, Lower Costs

Complex FDC modeling can involve Ph.ds with domain expertise, while deep learning can involve models with “30-40 lines of Python code,” he said, noting that the “minimal number of lines of code translates to lower costs.”

Humans, including engineers, are not adapted to look for small details in hundreds or thousands of metrology images or SPC charts. “Humans don’t do that well. Engineers still see what they want to see. We should let computers do that. When it comes to wafer analysis and log files, it is getting too complex (for human analysis). The question now is: Can we leverage these advances in machine learning to solve our problems?”

After training a model to detect distortions for a particular stepper, based on just 35 lines of Python code, Redman said the model provided “an extremely good match between the predicted values and the actual values. We have a model that lines up exactly. It is so good it is almost obscene.”

Redman said similar models could be applied to make sure etchers or electroplating machines were performing to expectations. And he said models can be continuously trained, using incoming flows of data to improve the model itself, rather than thinking of training as distinct from the application of the system.

“Most people talk about the training phase, but in fact we can train continuously. We run data through a model, and we can feed that back into the model, using the new data to continuously train,” he said.

Machine Learning for Predictive Maintenance

Benjamin Menz, of Bosch Rexroth (Lohr am Main, Germany), addressed the challenge of how to apply machine learning to predictive maintenance.

To monitor a machine’s vibration, temperature threshold, power signal, and other signals, companies have developed model-based rules to answer the question: Will it break in the next couple of days? Metz said

“Machine learning can do this in a very automatic way. You don’t need tons of data to train the network, perhaps fifty measurements. A very nice example is a turning machine. The network learned very quickly that the tool is broken, even though the human cannot see it. The new approach is clearly able to see a drop in the health index, and stop production,” he said.

3D Chips, New Packaging Challenge Metrology and Inspection Gear

Monday, March 21st, 2016

Compiled and edited by Jeff Dorsch, Contributing Editor

Metrology and inspection technology is growing more complicated as device dimensions continue to shrink. Discussing crucial trends in the field are Lior Engel, vice president of the Imaging and Process Control Group at Applied Materials, and Elvino da Silveira, vice president of marketing, Rudolph Technologies.

1. What are the latest market trends in metrology and inspection?

Lior Engel, Applied Materials: The market trends we are witnessing today are influenced by the memory mix growth in wafer fab equipment and emergence of technology inflections as the industry progresses to advanced nodes and 3D device architectures. The optical inspection market is growing along with wafer fab equipment. We have seen the memory mix of wafer fab equipment grow from 23 percent in 2012 to almost 50 percent in 2015. The memory growth trend along with the transition from planar to 3D NAND changes the dynamics, as 3D NAND in general requires more metrology solutions while the foundries are maintaining high demand for optical wafer inspection. Demand for electron-beam products is increasing for all device types.

Shrinking design rules and shrinking process windows translate to systematic defects becoming a critical issue. These can hinder time to yield and affect production yields. The interaction between the design and process can fail under certain process conditions, and the resulting small defects are extremely difficult to find. Challenges such as these are fueling the need for both optical and e-beam inspection solutions in the fab. These different solutions complement each other and help the fab throughout the entire chip lifecycle. From a market perspective, the e-beam inspection market continues to grow and outperform WFE. E-beam inspection is currently focused at the R&D stage but is beginning to shift to high-volume manufacturing.

The metrology market is also growing due to multi-patterning requirements, the need for increased measurement points and tighter process window control. The advent of e-beam massive metrology tools provides a solution for process monitoring and uniformity control. Also driving the market are the ever increasing high aspect ratio (HAR) 3D NAND devices in memory and the increasing complexity in 3D FinFET metrology in foundry.

The workhorse metrology solutions include CD-SEM for multi-patterning controls and HAR memory, and optical critical-dimension (OCD) addressing spacer profile reconstruction in multi-patterning and full device characterization in FinFET.

Elvino da Silveira, Rudolph Technologies: In our experience, fan-out wafer level packaging (FOWLP) is a big trend for our customers. FOWLP does not require a substrate, so the lower cost makes it an attractive packaging technique over 2.5D or embedded interposers. There is a wide range of low- to high-end FOWLP applications, such as MCP/SiP, PoP, and 2.5D FOWLP, each requiring specific inspection/metrology techniques.

Further, we see submicron inspection as a big trend fueled by shrink. More than Moore is driving creative packaging that requires inspection of shrinking redistribution (RDL) lines. Miniaturization and multiple functions packaging, driven by the wearables and Internet of Things market, creates more emphasis on microelectromechanical system (MEMS) devices and sensors. Also, shrinking nodes in the front-end have shifted macro inspection needs to the submicron level.

With regards to front-end metrology trends, 3D is the driver. Second- and third-generation FinFET and 3D memory (both DRAM and NAND) are the key market drivers for front-end logic and memory.  We are also seeing radio-frequency (RF), MEMS, and CMOS image sensors (CIS) move to adopt the latest generation of metrology as they compete to improve their processes and gain market share.

2. What are the latest technology trends in metrology/inspection?

Lior Engel, Applied Materials: Inflection challenges that are affecting M&I technology trends include:

 Design rules shrinking causing denser feature imaging and the advent of smaller killer defects

 3D transistors having more complex geometries, trenches and sidewalls. There is no line of sight to the killer defects and new materials are being introduced

  •  HAR structures introducing buried defects and new metrology challenges
  •  Process marginality resulting in critical systematic defects that require metrology coverage

In optical inspection, the technology trends addressing growing challenges include sensitivity improvement by enhancing the signal from key defects of interest. This can be achieved by enhancing imaging techniques and nuisance separation capabilities. In addition, leveraging design information (CAD) and optical information to optimize nuisance filtering.

For e-beam applications, which include SEM review, CD-SEM metrology and e-beam inspection, trends include:

 Adding on-tool automatic classification and analysis capabilities, which result in more meaningful statistical process control (SPC) and yield control. Automation produces faster enhanced results, reducing human error and speeding up the process

 Achieving 1-nanometer e-beam resolution and the availability of new imaging techniques are being utilized for finding the smaller defects in complex structures

  •  e-beam massive critical dimension (CD) measurements are being used for uniformity control
  •  e-beam voltage contrast inspection is increasingly required for embedded defects in 3D structures
  •  In-die on-device 3D and overlay measurements are challenging current optical metrology techniques, and trending towards new in-line solutions such as e-beam.

Elvino da Silveira, Rudolph Technologies: Increasingly complex front-end processes paired with “More than Moore” advanced packaging techniques are resulting in die-level stress. Product loss at assembly is extremely expensive since it’s one of the last steps in the process. Singulation excursions can manifest as a yield problem, but most often result as a reliability problem making them harder to detect and control. Traditional automated optical inspection (AOI) has been focused on active die areas rather than total chip area and is somewhat difficult and prone to overkill. Rudolph has developed a method to detect and monitor wafer chipping without extra investment or tool process time.

We solve the AOI inspection challenges by specifically monitoring the die seal ring while simultaneously inspecting both the active area and remaining kerf area, avoiding any throughput penalty. With our high-sensitivity/low-noise pattern-based inspection, customers can decide how close chips can occur relative to the seal ring. Judgements can be made about die quality based on certain characteristics (distance between die, die rotation, etc.). Lastly, customers can review image capture in both visible and infrared (IR).

Another inspection technology trend we see is the need for detection of non-visible/low-contrast killer defects in 3DIC flows. A 3D stacked IC flow may require a combination of through-silicon via (TSV) formation followed by die-stacking and molding. TSV interconnect formation flow will require processes such as via etch, via fill, nail reveal, copper pillar, wafer bonding, and debonding. A comprehensive process control strategy for such a complex flow requires multiple inspection and metrology approaches. Bright-field and dark-field detection is the baseline inspection technology for random and systematic defects. As the processes for TSV take on a more fab-like look, and are implemented in what is now being called the middle end, attention is turning to defects that are normally not visible. Examples of non-visible defects range from voids in TSVs to faint organic residues and incomplete etch on the bump pad. Voids can be detected using laser acoustic metrology. Laser acoustics also offer a unique solution for measuring the individual layers in a pillar bump stack to ensure tight process control and device yield. Organic residue-based defects have been tedious to detect using manual fluorescent microscopes. Now a more reliable approach to detecting organic defects is possible using automated high-speed fluorescent imaging based inspection. The strategy of combining bright-field, dark-field inspection with automated fluorescent imaging inspection, laser acoustics and software to analyze defect and metrology data has proved to be a cost effective approach to managing visible and non-visible defects in advanced assembly flows.

Advanced patterning of three-dimensional gate structures and memory cells is driving the need for advanced metrology techniques. Some of which have not been developed yet! Optical CD, X-ray, and acoustic metrologies are all at the leading edge. Optical wavelength ranges are now upwards of 20 microns to deal with thick multilayer memory stacks. Missing layer detection and the ability to measure ultrathin metal stacks with complicated interface characteristics are also challenges faced by our customers.

3. How are equipment vendors helping find defects in the nanoscale era?

Lior Engel, Applied Materials: Vendors must combine enhanced resolution, advanced imaging, and smarter applications into their offerings to meet the increasingly complex requirements from chipmakers as they transition to advanced nodes and 3D devices. E-beam and optical inspection solutions must become faster and more sensitive.

Metrology solutions are being used beyond traditional systematic process control, generating massive high-sensitivity data that is leveraged for predictive analysis.

In addition, as challenges grow, advanced applications leveraging design data and machine learning capabilities improve the overall results that the tools can deliver.

Elvino da Silveira, Rudolph Technologies: Those suppliers that can not only provide the required technology, but also provide the ability to take multiple points of data from across the fab, analyze that data, and make it actionable. True end-to-end process control that reduces time-to-ramp and improves ramp to yield—this is the value proposition that Rudolph offers its customers.

4. How is the 2016 market shaping up?

Lior Engel, Applied Materials: As was stated in Applied’s latest earnings call, our market outlook, taking into account the global economic climate, is that wafer fab equipment spending levels in 2016 will be similar to 2015. Driving industry investment are the technology inflections around 10-nanometer and the shift to 3D NAND, as well as increased spending in China.

Elvino da Silveira, Rudolph Technologies: Although Gartner is forecasting a flat 2016, Rudolph is uniquely positioned in both the front-end and true back-end semiconductor processes in a number of growth markets. Additionally, our new product pipeline is strong.

We see an opportunity to outperform our peers in 2016.

5. Is business improving, declining, or staying flat this year?

Lior Engel, Applied Materials: While the overall spending trend for WFE this year is flat, we are maintaining a positive outlook for Applied in 2016 because our customers are making strategic inflection-driven investments that play to our strengths. Our position is optimistic on wafer inspection for 2016. Our latest UVision Brightfield tool has a good position in foundry and logic. We’re the leader in e-beam review and are now taking that technology into inspection where we have significant pull from customers. So I think overall in 2016, we’re pretty optimistic about that business.

Gases: Completely Necessary to Semiconductor Manufacturing

Wednesday, July 15th, 2015

By Jeff Dorsch, Contributing Editor

Silicon wafers. Semiconductor packaging. These are commonly known products in materials for manufacturing and assembling microchips.

Process gases? Not as familiar a commodity, yet they are just as critical to semiconductor manufacturing.

Some of the biggest gas suppliers have reported greater profits this year, while revenues have declined for certain companies, largely due to the strength of the U.S. dollar against other currencies.

Air Liquide reported revenue in its Gas & Services segment was up by 6.3 percent in the first quarter, while its Electronics grew revenue 14.4 percent compared with a year ago. “Sales were particularly vigorous in China, in Taiwan, in Japan, and in the United States,” the company said in a statement.

For its fiscal second quarter ended March 31, Air Products & Chemicals reported net income was up 19 percent from a year earlier to $336 million, while sales declined 6 percent to $2.415 billion.

“Volumes increased four percent, primarily in Industrial Gases-Asia and Materials Technologies, and pricing was up one percent,” the company stated. Industrial Gases-Asia had a 7 percent increase in sales during the quarter, to $393 million. “Electronics Materials sales were up 16 percent on strong volume growth in all business units and positive price,” Air Products said.

Praxair reported first-quarter net income of $416 million. Sales were $2.757 billion, down 9 percent from a year earlier, due to the impact of negative currency translation. The company is forecasting 2015 revenue of $11.4 billion to $11.7 billion.

Financial results from the quarter ended June 30 will be reported later this month.

“The quantity and number of gases are increasing quite dramatically,” says Anish Tolia, head of global marketing for Linde Electronics, part of the Linde Group. “Fabs are getting larger, clustered in one location. The amount of gas per wafer is increasing.”

Some of the increases in gas usage are due to double patterning and multiple patterning in lithography steps, according to Tolia. “More tools, more gas,” he observes.

Nitrogen is the most-used gas, typically employed for purging of pumps and vacuum chambers, Tolia notes. “It is also used as a process gas,” he says. Wafer fabrication facilities making chips with 28-nanometer or 20nm features can go through 20,000 to 30,000 cubic meters in an hour.

Hydrogen is another significant commodity gas, Tolia says. “Extreme-ultraviolet lithography will boost its use,” he adds.

Process gases are in demand for etching and deposition, particularly in chemical vapor deposition, and less so in physical vapor deposition, according to Tolia.

The semiconductor industry is a “strong consumer” of helium, a cooling gas, Tolia says. “Supply has improved,” he notes, with the U.S. Bureau of Land Management maintaining a helium reserve. “A few new sources have been developed; the pressure is off,” he adds.

Helium, which comes with natural gas, is not a renewable resource, Tolia notes. “It’s not a gas from air,” he says.

With the concentration of fabs at some locations, such as the Hsinchu Science and Industrial Park in Taiwan, those sites could have a generating facility. “We could set up an air separation unit near there,” Tolia says. And some fabs have captive air separation units.

With on-site supply, “as close to their fabs as possible,” Linde Electronics is better able to serve its customers, Tolia says, with “minimized leadtimes, quick troubleshooting, quick response to disasters.”

Gases are essential to the semiconductor industry. Have you thanked a gas supplier lately?

Solid State Watch: June 12-18, 2015

Thursday, June 18th, 2015
YouTube Preview Image

Extension Media websites place cookies on your device to give you the best user experience. By using our websites, you agree to placement of these cookies and to our Privacy Policy. Please click here to accept.