The irony of the week: Qualcomm, leader of the wireless world, sends a large contingency to participate in and sponsor IMAPS 2012, taking place at the Town and Country Resort and Convention Center in San Diego (I use the term “resort” very loosely), and the wireless was crap (really, I tried a few other “nicer” words to describe it, but there’s no better way.) When you’ve got a hotel full of engineers from the microelectronics industry, this is NOT a good thing. Really, it was all anyone was talking about on the first day, which is a shame, because IMAPS worked hard to put together a strong program. Applied Materials’ Sesh Ramaswami suggested a simple solution when he noticed a line in the registration area: an 11g access point router should do it, and what a sponsorship opportunity for Qualcomm! Luckily, by the second day, connectivity is markedly improved (presumably without taking Sesh’s advice), which is why I was able to finish this post.
One thing’s for sure, it was a great real-life illustration for Matt Grob’s (CTO, Qualcomm) keynote address, which discussed how to increase performance of cellular tech in wireless networks 1000x. His talk was pretty high-level, explaining how in the US, only 500MHz of the RF spectrum is available for mobile applications, which is really not enough (clearly!), and how Qualcomm plans to increase 10x over 10 year period.
One way is to shrink base stations and place numerous small cells around to increase the capacity across a metropolitan area. He explained the concept of an “inside out” model vs. the current “outside in” model. 70% of traffic all wireless data traffic is indoors. The “outside in” model has shortcomings; the signal needs to penetrate walls, etc. Qualcomm has discovered you can improve performance by putting small base stations indoors, and if enough users deploy these indoor base stations, you get decent coverage outside as well. “I’m not suggestion you can make an entire network this way, but it can be an excellent supplement to the macro.” said Grob. The company is focusing on reducing small cells cost and size. The size of a cell phone, these mini base stations already cost less.
Another new idea for cellular Qualcomm is working on is device-to-device communication where cell phones “talk” directly to each other, rather than bouncing off a cell base station first. Grob says this will enhance proximal communication. They call this technology LTE-Direct. Here’s an explanatory video:
Qualcomm’s Approach to Wide I/O Dram on Logic
Qualcomm has invested a lot in 3D IC technologies (perhaps to enable the aforementioned devices? It wasn’t specified, but one can draw conclusions). I spoke with Sam Gu, who presented a prize-winning paper (best written of show) on the company’s wide I/O DRAM on Logic and their vision of the supply chain that would manufacture this. The objectives of this work, Gu told me, was twofold: verify that a micro pillar gate array (MPGA) wide I/O memory cube can be stacked on 28nm CMOS logic; and verify a supply chain model similar to the Package on Package (PoP) supply chain.
The first objective was achieved. “There’s no technology show stopper to enable this technology for wireless applications. High yield was achieved on champion samples.” noted Gu. However, he added what’s required to enable this technology for high volume manufacturing (HVM) is a low MPGA price and a favorable business model.
To this end, Qualcomm’s supply chain vision is for the foundry to do the TSV processing, and the OSAT to do assembly processes. Currently they are handing off the thin wafer on a film frame, but they’re not opposed to the thick wafer handoff. The challenge is that for OSATS to handle the backside processing requires some learning to make sure they don’t disturb the advanced CMOS devices, explained Gu.
To make it more affordable, Qualcomm is promoting a PoP- style business model, in which the handset manufacturer procures the logic die from the fabless/foundry, and the memory from the memory vendor, and then the Wide I/O DRAM on logic chipset is assembled at the OSAT. In this scenareo, each vendor holds the liability for the known good die (KGD) or in the case of memory, the known-good stack (KGS). Reliability failure due to assembly would be owned by the OSAT. Gu noted that the OSAT needs to strategize and have a certain budget for assembly failure. There’s not going to be 100% yield, he said.
The company’s reasoning for this model, rather than procuring the memory themselves, is a reduced system cost. Qualcomm does not design or manufacture memory, so if they handle procurement, they need to add on a profit margin, explained Gu. Therefore it’s a cost savings for the handset manufacturer to handle it themselves. “The current PoP model is an established business model, so it’s not a stretch.” he noted, adding that the big difference is that PoP is dealing with stacking complete tested packages, while with logic on memory, we’re dealing with chipsets, which are more delicate and come with a higher liability. The key lies in the known-good factor, and making sure as much failure is uncovered prior to assembly. ~ F.v.T.