After attending last week’s 9th Annual Architectures for Semiconductor Integration and Packaging Conference in Redwood City, CA, it’s pretty clear that there’s probably the same chance of 3DIC NOT happening, as there is a chance the world will end on Dec 21. (And oh by the way, Tom Pawlowski, chief technologist and fellow, Micron Technology, says he’s having a Day After the End of the World party on the 22nd and we’re all invited.
What makes me so confident? For one thing, despite all the remaining challenges (and make no, mistake, there are some BIGGIES left that keep pushing out the roadmaps to adoption), the pain threshold for a solution is rising. As Pawlowski explained to me sagely, “3D is the solution for a situation that has no solution.” As such, challenges aren’t mentioned as reasons to give up the effort, but as things that MUST be addressed so 3D can proceed. We have no choice but to figure it all out, because pretty soon, the other alternatives that are available will no longer be sufficient.
The FPGA manufacturers have already embraced silicon interposer technologies (2.5D or 3D, depending on who you talk to) wholeheartedly. In responseto the all-encompassing question “Are we there yet?” Arifur Rahman, Product Architect for Altera, Corp. said in his welcome comments, “It’s safe to say we’re in the middle of it and have product realization.” He added that we are approaching the goal and excitement is building. “There are lots of announcements, and everybody’s roadmaps are coalescing,” he noted. “Now we have to work on bringing the price down so it can be used in the mobile area.”
“The only way this will be come a pervasive technology is if a number of people play, and thus bring down the cost associated,” noted keynoter Liam Madden, corporate VP of Xilinx. Xilinx has been shipping product to customers for nine months. Madden talked about his experience with 3 decades of microprocessor Integration, and how he believes that advanced high performance heterogeneous products will be best achieved through die stacking. “The economics of stacking has come along and whacked us on the side of the head,” noted Madden. “We’re interested in building products today, and today, scaling doesn’t make economic sense.” He dubbed this theory, “Stackonomics.”
Madden also talked about closing the “packaging chasm” by getting rid of the packages and opting for heterogeneous chip integration. For Logic, Memory and Analog, using through silicon via (TSV) interposers solves the “more bandwidth at lower power” problem, because it allows for close intergration and thousands of interconnects. Xilinx flagship interposer-based product, Virtex-7 is enabling in one device what normally takes more devices to do.
In his concluding remarks, Madden noted that while we can’t have technology at any cost, he doesn’t think this 3D is going to break the bank. “Look at system cost,” he advised, “cost is an interesting thing. It can’t be just a part cost against part cost.” He also said that rather than wait for all the challenges to be ironed out, we can make significant progress with existing technologies, and learn something along the way.
Listening to the presentations of Dave McCann of GlobalFoundries, and Paul Silvestri of Amkor Technology, its difficult to NOT get caught up in the enthusiasm that is reaching almost a fever pitch among the foundries and OSATS. The overall message: We’re ready! So somebody design this in and lets get on with it.
McCann’s message focused on GlobalFoundries collaborative supply chain model. He said the company supports wafer fab through bump and probe, and would ship full thickness wafers to the OSATs for backside processing. The foundry has been working closely with Amkor to establish test points on its 2.5D work; Amkor does the thinning, final assembly and test. McCann said the company is putting all its findings into process design kits (PDKs) so that design rules can be established.
“The development of a flexible supply chain is working,” reported McCann. “The partners are generating lots of reliability data, A flexible supply chain may be difficult to implement, but will provide optimized solutions to customers so that they can offer varied solutions based on applications.” McCann expects that the company will be ready for commercialization when customers are: 2.5D interposers by the end of 2013 and full 3D IC by 2015.
Silvestri, presenting on behalf of Ron Hoemueller, provided status reports on Amkor’s capabilities. Essentlally, he says Amkor is ready to roll on all process steps except for wafer support systems (WSS) for thin wafer handling. They’re also actively pursuing thermal management solutions. “We believe we’re ready to move forward,” says Silvestri. “We’re meeting reliability requirements. It’s a really good story from ten years ago. We’re passing harsh reliability milestones.”
But the hard truth is, challenges remain, and as long as there are reliable, cost-effective alternatives that meet performance needs at a lower cost, the wait for 3D in mainstream, high-volume manufacturing (HVM) continues. TeachSearch’s Jan Vardaman, who can always be counted on to bring us back to earth, advises patience. Yes, Wide I/O promises to provide twice the bandwidth ass DDR 2 at the same power, and TSV solutions offer the most compact, thinnest, best performing potential, but “Roadmaps shift out because we still have obstacles to work on. You have to give this stuff time to develop,” she says. “When Xilinx first started, they didn’t know what they didn’t know.” It’s taken 6 years of commitment and investment to bring them where they are today. “We have to have a realistic view of the timeline associated with this.” In the mean time, she says we’ll continue to see novel ways to extend current technologies until we get to 3D ICs. Examples of this include Amkor’s TMV-PoP, embedded die packages, and fan-out wafer level packages (FO-WLP) like eWLB. Others such as Invensas’ xFD and BVA technologies, Deca’s M-Series, and Amkor’s Possum also come to mind.
But ultimately, no about of rain will dampen this parade. You could feel it in the air, and hear it in the tone of presenter’s voices as they updated attendees on everything from design tools, design for test methodologies, and thermal management, to process fine tuning for cost reduction, and of course the ever-elusive temporary bond/debond conundrum. The weight in emphasis has shifted from “can it be done?” to “let’s get it it done!”, and big companies are finally on board and driving 3D adoption. Challenges will be met. 3D will happen. Patience will be rewarded.