Engineers are finding that using the cloud for data storage, analysis, and instrument control offers an alternative to using internal resources

BY MARTIN
ROWE, Senior Technical Editor, Test & Measurement
EDN and EE
Times
www.edn.com, www.eetimes.com

When you
think of “the cloud” and “big-data analytics,” you probably think of a nebulous
place where marketers store data about everyone’s whereabouts and buying
preferences. But wait, engineers don’t do analytics; they do analysis.
Nevertheless, having test data stored and processed in the cloud can offer some
advantages over storing it locally.

Several
companies have developed web-based apps for using the cloud to store test data,
analyze that data, and even control bench instruments. So far, each application
we’ve seen serves a different purpose.

Why store
test data in the cloud? There are several reasons, but they come with
tradeoffs. Advantages include:

  • Dynamically change computing power and
    storage capacity.
    You don’t
    have to go through the process of getting more power and storage from your IT
    department.
  • Share test data with colleagues, suppliers,
    and customers anywhere in the world.
    Think of it
    as a Google spreadsheet for test data but with far more analysis power.
  • Monitor product test status from your desk or
    while traveling.
    You don’t
    have to wait for someone across an ocean to send test data.
  • Analyze large amounts of data and look for correlations between
    measurement data and metadata (test conditions) on the server, keeping your
    computer, tablet, or phone free from any data processing.

     

On the
downside, data is likely more secure if stored on your company’s internal
network. That’s a choice you’ll have to make when deciding to use a third-party
cloud service or store and process data on internal servers.

San
Francisco-based company GradientOne has developed a cloud-based app that lets
you automate test benches in the lab or on the manufacturing floor. It’s the
only app we’ve seen that can control hardware directly. Using a hardware
gateway between the internet and your test equipment, GradientOne’s app uses “drivers”
designed to control oscilloscopes, meters, and spectrum analyzers. You can
automate tests using an application programming interface (API) or from a
soft-front panel.

GradientOne’s
web-based app lets you move test data directly from test instruments to the
cloud. Alternatively, you can collect data offline and then upload to GradientOne’s
cloud-based app for data storage and analysis. Because test data can come
directly from your test equipment, you can not only generate reports, but you
can see specific test results, including waveforms (Fig. 1).

0517_ProdTrends_Fig1_GradientOne_RGBcurve

Fig. 1: A cloud-based app from
GradientOne lets you control test instruments such as oscilloscopes, display
waveforms, and process measurement data. Source: GradientOne.

JitterLabs,
based in Milpitas, California, has developed software designed to help you
select clock oscillators for high-speed serial interfaces. The problem stems
from the fact that serial-data specifications have differing requirements for
bit jitter. Plus, there’s no established test method for measuring jitter and
noise in clock sources.

JitterLabs
has developed a measurement method that the company offers as a service. A cloud-based
app takes data from measurements taken at JitterLabs and stores the data on two
servers — one stores the data and the other performs the data processing. The
servers, owned by JitterLabs, are hosted by a third party. Jitterlabs’ Gary
Giust notes that the server used to store the data uses hardened security. The
database stores measurements and metadata about each measurement. For example,
data might include the manufacturer’s name and part number for each reference
clock, the oscilloscope used to measure jitter, and specifications from clock
manufacturers such as phase noise. Once the clock signals are stored, you can
process the data using filters and compare the results against published
standards. The graphic in Fig. 2 shows where
data comes from and what the cloud-based app produces.

0517_ProdTrends_Fig2_Jitterlabs_ecosystem_1

Fig. 2: JitterLabs uses a
cloud-based service to filter data from clock sources, from which you can
choose a clock for high-speed designs. Source: JitterLabs.

Keysight
Technologies has introduced its N8844A cloud-based software visualization tool.
It helps engineers and managers review test results and make decisions based on
those results. Keysight’s R&D Manager, Brad Doerr, explained how engineers
use the N8844A’s data-analysis tools to track projects and analyze data by
using the cloud’s processing power. You can plot data based on measurements
versus a specific parameter, such as bit errors versus input voltage or output
power versus temperature. Fig. 3 shows a
histogram of bit errors.

0517_ProdTrends_Fig3_Keysight_N8844A

Fig. 3: The N8844A from Keysight Technologies
lets you process and share test data with colleagues. Source: Keysight
Technologies.

Semiconductor
manufacturing generates tons of data. With engineering and manufacturing often
separated by oceans, having a way to easily share data and analyze test results
can shorten problem-solving times. Optimal+ software for manufacturing lets you
store data in the cloud, share it, and analyze it. Fig. 4
shows how you can use scatter plots to identify potential recall parts before
they reach customers.

Optimal+’s Director
of Marketing, David Park, explained that when returned parts exhibited certain
characteristics in final tests, a manufacturer looked through test data and found
other parts that differed from the norm. Once the correlation was made between
this particular test result and returned parts, engineers created a rule that
flags parts whenever these conditions occur, thus preventing potential questionable
parts from being shipped.

0517_ProdTrends_Fig4_Optimal_outliers

Fig. 4: Software from Optimal+
runs in the cloud or on local networks, collecting semiconductor manufacturing
data for analysis. Source: Optimal+.

Another
cloud-based app for semiconductor manufacturing comes from MFG Vision, which is
based in Limerick, Ireland. MFG Vision’s software is aimed at fabless
semiconductor companies who use third-party foundries and test houses. The
company has developed a series of cloud-based apps that let you monitor
manufacturing and test activity in real time.

The company
provides scripts that also let you upload data from your company’s operations
or from third parties. Because MFG Vision provides these scripts, the company
guarantees that test data will be uploaded to the cloud, said company founder,
John O’Donnell. Data storage and processing can be hosted on a third-party
cloud service or on a company’s internal network.

Once data is
uploaded, metadata is extracted from the raw data. The server processes data,
letting you add filtering and correlation that looks for conditions that might
lead to failures. Test data can be from assembled ICs or from on-wafer devices.
Fig. 5 shows yield data for a lot of 77
wafers. Colors indicate the yield percentage for each die location.

0517_ProdTrends_Fig5_MFG_wafermap

Fig. 5: One of several
cloud-based apps from MFG Vision produces a yield map from a set of wafer
tests. Source: MFG Vision.

In addition
to providing data analysis, MFG Vision’s apps let you collaborate with
colleagues so that anyone with login privileges can share data and plots. Plus,
anyone can perform analyses on the data because it resides in the cloud.

Conclusion

Regardless
of how you use cloud-based (or any other) measurement and analysis software,
your results are only as good as your measurements and the data you collect. Keysight’s
Doerr noted that metadata should include not only physical conditions placed on
the device under test (DUT) — power-supply
voltage, temperature, etc. — but it
should include information such as process variables, who ran the tests, and
with which equipment. You may find that experienced engineers get better test
results. If one test setup produces data different from others, then something
may be out of calibration. The software or firmware installed in your DUT can
make a difference, so track that, too.

All of the products
covered here can process data and can let you correlate test results with test
conditions. Look at them carefully and evaluate them against your needs.