The Changing Role Of The OSAT

By Ann Steffora Mutschler
As process geometries and packaging technologies have matured over time, the OSAT (outsourced semiconductor assembly and test) provider has played an evolving role in the semiconductor packaging ecosystem. With true 3D chip stacking on the horizon, their role may evolve once again as ecosystem players jostle for position in the 3D universe.

There are two things that are likely to change with 3D, according to Raj Pendse, vice president and chief marketing officer at StatsChipPac. “The first one is that the very thought that a foundry would look at 2.5D as an extension of their SoC integration/monolithic silicon approach would mean that they would try to control certain parts of this flow.”

The foundry would then try to convince the end customer to direct the memory chips to them so they could integrate them at the interposer level, because many times what they are doing—like Xilinx—is partitioning the chip into multiple chips.

“They used to make the monolithic chip,” said Pendse. “Now they make the four daughter chips, and then they integrate that heterogeneously with memory or something else. In order to keep the value of the silicon in their control, they would want to do the integration. The whole concept that that would simplify the transition to new silicon nodes would only work if they control the integration. That’s why they are very stubbornly trying to keep control of that. That would change the model because then the integration function would move from the OSATs to the foundry. That might actually happen.”

However, Pendse believes those models will coexist. “It’s not going to be all or nothing. There will be a certain percentage of cases where OSATs would still be the integrator and other percentage would be the foundries. That would be an additional element of who is being the integrator.”

Ron Huemoeller, senior vice president for advanced 3D interconnect platform development at Amkor agrees that business logistics definitely will change as a result of 3D—in part because there are many different entities and heterogeneous materials that all need to go back into an assembly format. “It used to be pretty clear who owned what in the past, for example, memory ownership, logic ownership, die wafer ownership in general, because all that was finished and shipped as a completed entity to OSATs to just die and stack. That is no longer the case. Because of that, the responsibility will shift with that.”

In many cases where logic is being finished at an assembler’s house—whether an OSAT, IDM or a foundry like TSMC—it changes the responsibility of the quality and outcome of that wafer. When responsibility is shared, there is not necessarily a metric in place today for that handoff, he asserted.

“Most likely what’s going to happen is the original exchange methodology will stay in place with the caveat of packagers only in back-side responsibility and then in the wafer yields, as a result,” said Huemoeller. “Wafer mapping becomes very important at this stage to identifying known good die at the wafer level. There will be a bigger burden on the wafer foundry people to do very good job with wafer mapping so we don’t process and take bad die and move them further on.”

OEMs regaining dominance?
A second trend StatsChipPac sees today is the probability of the OEM gaining dominance in the ecosystem. “As you see today, if you look at the mobile phone market the dominant players really in tablets and smartphones are Apple and Samsung,” Pendse said. “They both design their own chips. They even detail the design of the packaging. They sometimes, in the case of Samsung, completely captively produce everything. We’ve never had a business relationship with OEMs until now because we sell packaging services to intermediate companies like Qualcomm. We never talk to Nokia and we don’t sell anything to Nokia, HBC or Cisco. But we think that will probably change, as well. There is already a trend now for guys like Apple and Samsung to directly work with silicon and package foundries. So then they will do even a higher level of integration because, from their perspective, they can provide the memory as well. And they may find it more effective to integrate at the OSAT level.”

He believes there two types of companies that will engage in this way. One involves dominant companies such as Nokia. The other type would be the China-based OEMs because, as Pendse explained, “in China the market is un-penetrated still and the strategy of China is to nurture their own ecosystem and there are huge, huge OEMs—Huawei, ZTE and Lenovo—and they are already doing this. They are starting to control the supply chain directly and they have brought up their own design companies like Huawei has brought up HiSilicon, and they are dictating the packaging, etc. This model is very easy for them to adopt and it actually will save some money because they will eliminate the middleman.”

From the user perspective, Bob Patti, chief technology officer at Tezzaron Semiconductor, who has been in the trenches 3D-wise, pointed out that at the end of the day, the other thing which seems to rule out the TSMC model of the world is that Tezzaron takes wafers from different fabs. He believes this will be the norm.

“We have to have materials information to know what kind of materials are in the stacks and we usually need to know specific stack thicknesses and stack ups,” Patti said. “The fab guys would really like to know if they break a wafer somewhere, what have they done? What materials have they exposed that piece of equipment to?”

This is where things get tricky from the foundry perspective. A foundry might be willing to give a Tezzaron the information. But consider this: “Nintendo’s going to build their next-generation box,” said Patti. “They get their graphics processor from TSMC and their game processor from IBM. They are going to stack them together. Do you think IBM’s going to be real excited about sending their process manifest—their backup and materials—to TSMC? Or maybe TSMC will send it IBM. Neither of those is ever going to happen. Historically they do share this with OSATs, or at least some material information. And they’ve shared it typically with us because I’m only a back-end fab. I can keep a secret. I’m not going to tell their competition. There’s at least a level of comfort in dealing with a third party that isn’t a competitor.”

This is why the fabs end up not doing the 3D work, he believes. “They end up in the position which ultimately compromises the other material that needs to be brought in.”

At the end of the day, however, the biggest 3D impact to the OSAT is collaboration. As Huemoeller puts it: “We’re seeing a whole new paradigm shift in collaboration with our customers. It used to really be a top-down product development or business environment. Today, it is very collaborative from the beginning stages of their new product platforms where we are involved form the initial stages in their design now and working side by side with them to get products out to the market. This is very new. Typically we didn’t get involved at the very early stages of platform planning stages. We do with 3D.”