Midtown Blogger/Manhattan Valley Follies: Extreme Tech: Samsung and Espionage

Sunday, February 8, 2015

Extreme Tech: Samsung and Espionage

EXTREMETECH
COMPUTING
DID CHIP ESPIONAGE, IP THEFT GIVE SAMSUNG ITS 14NM MANUFACTURING LEAD?

Did chip espionage, IP theft give Samsung its 14nm manufacturing lead?

By Joel Hruska on February 6, 2015 at 1:02 pm
33 Comments

Share This Article

For most of the past decade, the semiconductor industry has been led by the same pair of companies: Intel was the most-advanced integrated device manufacturer (IDM) and led the entire market, while TSMC was the most advanced contract manufacturer (sometimes called a pure-play foundry). Samsung’s leapfrog over TSMC to become the 14nm provider of choice. A recent article has argued that Samsung’s sudden change in fortunes was no fortuitous accident, but the result of sniping several critical TSMC employees. One man in particular is alleged to have boosted Samsung’s efforts — Liang Mong-song.

TSMC first sued Liang back in 2011, alleging that the chip designer had given trade secrets to Samsung and broken his non-compete agreement. As evidence, the company submitted a lengthy report it commissioned from outside experts which compared various features of TSMC products against the manufacturing characteristics of their Samsung counterparts. As Samsung moved to lower process nodes, TSMC argued that its products began to increasingly resemble TSMC’s own hardware. According to TSMC’s engineers, the two foundries’ now have nearly identical 14/16nm process nodes.

According to TSMC, Samsung began using a similar diamond shape for silicon-germanium rather than IBM’s u-shape. Image courtesy of ChipWorks.

This isn’t as far-fetched as it might seem. Samsung has always worked closely with IBM (or had, up until the latter’s exit from the foundry business), and the entire point of the Common Platform Alliance between GlobalFoundries, IBM, Samsung, and Chartered (now part of GF) was to speed development and time-to-market by standardizing on a common set of methodologies. So, the fact that Samsung’s technology evolved to be less like IBM’s and more like TSMC’s could be construed as evidence of wrongdoing. Certainly there’s reason to think that Liang may have breached the spirit of his noncompete — apparently he left TSMC to teach in South Korea, but the foundry later discovered that the university he “taught” at was the Samsung Institute of Technology and his students were all veteran Samsung engineers. Liang was banned for working for Samsung until December 31, 2014.

The collision between trade secrets and individual knowledge

The judge in Liang’s case clearly felt that the engineer had engaged in a bad-faith breach of his non-compete agreement given that he was forbidden to work for Samsung for an additional period of months, but the punishment was a slap on the wrist compared to the potential damage to TSMC’s core business. According to Maybank’s Kim Eng:

When comparing to a full-node migration, ie 20nm to 14nm at Samsung and Intel, TSMC’s half-node approach 16nm underperformed in cost reduction (by as much as 25% if not higher), power consumption and performance. In a very rare case, intel infamously highlighted the potential risks of TSMC’s 16nm undertaking during its Nov-13 investors’ day. After the initial round of evaluation, many customers “strongly encouraged” TSMC to enhance its 16nm technology offering.

In other words, not only did Liang possibly tap his knowledge of TSMC’s cutting-edge implementations inappropriately, he may have done so at the worst possible time (from TSMC’s perspective). Samsung has come out of nowhere to lead in foundry manufacturing, at least in the short term. Maybank’s latest report on TSMC cut the company from “Hold” to “Sell” on the strength of Samsung’s 14nm ramp. According to Liang himself, he left TSMC after he was passed over for promotion and felt his work was under-appreciated by his former employer.

Tech ramps from TSMC, Intel, and Samsung

Yet Samsung’s leapfrog over its rival isn’t as simple as a simple guilty/not-guilty verdict. Of the several chip engineers we reached out to for insight on how trade secrets play out in the real world, all of them pointed out the same things — when you hire an engineer, you’re hiring them for what they know. If that engineer has an expertise in SOI, or FinFETs, or cutting-edge III-V materials, you’re obviously intending to tap that expertise. It’s easy to identify some types of IP theft, such as when an employee leaves with a trove of documents or confidential roadmaps. Situations like this are much thornier — TSMC, to date, hasn’t filed a lawsuit against Samsung or its former employee.

It may be tempting to pin Samsung’s rise to power on the actions of a single man, as CommonWealth magazine does, but I think this narrative is fundamentally inaccurate. Modern semiconductor manufacturing requires enormous capital input, teams of hundreds of engineers, and committing to a multi-year roadmap of iterative product improvement. No one person can singlehandedly drive this process for a sustained period of time. Whether Samsung’s 14nm lead turns into a sustained success or a momentary blip before TSMC retakes the pole position will depend not just on Liang, but on the entire ecosystem Samsung has built around its position and its ability to execute the contracts it takes now that its built the process node.

Now read: The semiconductor industry in 2015: New nodes and upstart manufacturers fuel a dynamic market

Hat tip to reader CoreDuo88 for the CommonWealth magazine link.