Applied Materials designs tools to leverage big data and build better chips

In semiconductor manufacturing, metrology — the science of measuring things — is an absolutely vital part of the manufacturing process. Much of this analysis is handled by CDSEM (Critical Dimension Scanning Electron Microscopy) equipment. As process nodes shrink and manufacturing difficulty increases, the amount of data being collected per wafer has increased. Foundries now collect more data per wafer than ever before, and they need to be able to analyze that information quickly and compare it to other readouts from different pieces of equipment. Applied Materials has launched a new web backend it calls TechEdge Prizm that’s designed to offer foundries better data on their day-to-day production and to do so in a far better manner than what’s currently available.

This image is drawn from an IBM 2013 SPIE paper from a study by Eric Solecky et al: SPIE 8681, Metrology, Inspection, and Process Control for Microlithography XXVII, 86810D (April 10, 2013); doi:10.1117/12.2010007.

With the amount of data per fab skyrocketing from 50TB per fab per year at 45nm to 80TB at 28nm, and an estimated 141TB at 14nm, better tools are needed for visualizing and examining system output closer to real-time. In the past, data was gathered by individual tools, locally stored, and painful to parse. There was no unified system for collecting information or comparing results between tools or across longer periods of time. With Prizm, Applied Materials hopes to change that. Instead of trying to parse data sets on a tool-by-tool basis, Prizm can gather data from multiple tools and present it through a unified interface. Results are searchable and can be analyzed much more quickly. Total time savings, again according to Applied Materials, are shown below.

The green bar is analysis time with Prizm, the brown bar is current time.

Prizm allows engineers to see various metrics on individual sections of a wafer map rather than simply as a chart of total data. Prizm is capable of showing how specific metrics have changed over time, or comparing specific metrics from one set of wafers against a later set. According to Applied Materials, Prizm can improve workflow efficiency by 10x in certain cases and spares engineers hours of tedious work manually gathering data. The online backend also stores data far longer — typical tools preserve data sets for a month; Applied Materials is guaranteeing seven years of storage for particular tools.

 

We spoke to Applied Materials about Prizm, and the company offered us a remote demo of how the service works. In the screenshot above, the engineer is able to drill down to examine metrics at each specific point on the wafer. Clicking on a section brings up an image of that area and gives more information on the selected metrics. The entire system is designed for flexibility — the engineers can examine and sort by tool type, process node, or a specific quality measure.

When Big Data matters

I’m skeptical of “big data” for the same reasons I’m skeptical of “cloud computing,” but the dramatic overuse and subsequent dilution of the latter phrase doesn’t mean there aren’t cases where cloud computing hasn’t offered something unique and different compared to the services we used to have. In this case, the  term “big data” term also seems to fit. Not only do these tools produce a staggering amount of information, the ability to sift and sort said research is essential to progress.

We’ve previously discussed the mind-boggling levels of accuracy the modern semiconductor industry requires as a matter of course, and the ability to measure those levels accurately is a necessity if products are to continue pushing below 20nm. Improving data collection and analysis doesn’t directly solve the problems facing the semiconductor industry, but it does ensure that the researchers working at companies like Intel, TSMC, and GlobalFoundries have access to the data they need to investigate defects more quickly.

Now read: Stanford builds first complex computer chip out of carbon nanotubes

• By Joel Hruska

Chinese physicists create first single-photon quantum memory, leading to quantum internet

A lab in China is reporting that it has constructed the first memory device that uses single photons to store quantum data. This is a significant breakthrough that takes us further down the path towards a quantum internet, and potentially quantum computing as well.

As it currently stands, we already make extensive use of photons — the bulk of the internet and telecommunications backbone consists of photons traveling down fiber optic cables. Rather than single photons, though, these signals consist of carrier light waves of millions of photons, with the wave being modulated by binary data. These pulses are never stored, either; when they reach a router, they’re converted into electrical signals, and then stored in RAM before being converted back into light.

A diagram showing the generation of a single photon (a), and the storage of a single photon with OAM (b)

Now, however, Dong-Sheng Ding and fellow researchers at the University of Science and Technology of China have announced that they have generated a single photon, stored it in a “cigar-shaped atomic cloud of rubidium atoms” for 400 nanoseconds, and then released the photon. The single photon is created using a process called spontaneous four-wave mixing, and the rubidium cloud stores the photon due to electromagnetically induced transparency (EIT). EIT causes a phenomenon called “slow light,” which is used here to “store” the photon for 400ns (more than long enough to count as computer memory).

The retrieved photon signal, vs. the storage timeThe photon, being stored

The generation and storage would be a big achievement in itself, but there’s more: the rubidium trap also preserves the orbital angular momentum (OAM) of the photon. As we’ve covered before, electromagnetic waves (including photons) can have both spin and orbital angular momentum. Spin angular momentum (SAM), which is equivalent to the Earth spinning on its own axis, produces polarization — and then there’s OAM, which is equivalent to the Earth rotating around the Sun. Generally, in wireless and wired communications, signals only use SAM and are therefore flat — but by introducing OAM, a signal becomes a 3D helix. You can encode a lot more data into a carrier wave  – perhaps an infinite amount – if you play with both the SAM and OAM. By preserving the OAM of the single photon, the Chinese researchers could be onto something very big indeed.

Moving forward, a photonic quantum memory is absolutely vital if we ever want to build a quantum internet out of quantum routers. Even if we pull back from lofty, quantum applications, if we could introduce OAM to the world’s fiber optic networks, the internet would suddenly get a whole lot faster.

Now read: Researchers create fiber network that operates at 99.7% speed of light, smashes speed and latency records

Research paper: arXiv:1305.2675 - “Single-Photon-Level Quantum Image Memory Based on Cold Atomic Ensembles”

• By Sebastian Anthony