Slashdot Mirror
For High-End CPUs, Qualcomm Ditches TSMC For Samsung
An anonymous reader writes: A report at Re/code says Qualcomm will have its next-gen Snapdragon 820 CPU made at Samsung's foundries, instead of TSMC's. The report points out a couple of good reasons for the switch: first of all, Samsung's plants run on a 14nm process, while TSMC still uses a 20nm process. Second — and more telling — Samsung recently ditched Qualcomm's Snapdragon processors for their new Galaxy S6 smartphone, opting to use their own Exynos chips instead. With the phone expected to sell upwards of 70 million units, that's a huge missed opportunity for Qualcomm. It's feasible Qualcomm could get Samsung to drop its own chips, because the Snapdragon 820 will have an onboard LTE modem. That would reduce the cost of assembling a phone, and also free up some space to make it smaller.
It's feasible Qualcomm could get Samsung to drop its own chips, because the Snapdragon 820 will have an onboard LTE modem. That would reduce the cost of assembling a phone, and also free up some space to make it smaller.
Yes, because the reason Android phones are so big is because they can't make the circuit boards smaller due to component count. It has nothing to do with stupid one-upsmanship on screen size.
This has nothing to do with Samsung, it has everything to do with tech process problems which spread well beyond the mobile. Case to point: desktop GPUs.
What is interesting is that some of the recent predictions about QCOM are not accurate. Lately QCOM stock has been heading down under two clouds. The first cloud is that the IEEE council appears to have sided with apple and other handset makers that the value of a FRAND license is not relative to the value of the phone (which has been the case up until now) but relative to value of the component that contains the patent. The rationale is simple: if the same component embodying the same patent could be in a cheap phone then the value should be tied to the common component. The latter is said to be about 1/10th the value of a high end phone. Thus QCOMs business model which includes royalties hit a snag. Thus it becomes vital that the components become expensive (and thus only be suited for expensive phones)---and the way to acheive that is to bundle up features into one uber component. Put the LTE in the processor. This is how they solve their problem.
The second reason the stock was headed down appears like it might be wrong information now. TMSC reported a huge drop in expected orders in the next year. It was known Qualcom was the cause of the reduced orders (well convincingly rumored). And thus it was assumed that qualcom was experience a loss of demand and reducing its TMSC orders. Thus the TMSC loss fed right into a QCOM stock price drop. But now we learn the drop was not QCOM reducing orders because of lack of demand for Qcom products but QCOM switching foundaries.
Since QCOM is switching boundaries so the TMSC part of the stock price drop was mistaken.
Finally, QCOM may be going upscale with a more costly process and going even more upscale by bundling patents into components. SO in one move they fix their business model problem. It makes sense too since broadcom has been targeting the low end SOC and wifi market where they can win on volume. Rather than compete at the low end where Broadcom could put Qcom patents in cheap hardware and pay little on the FRAND costs, QCOM veers luxury where it supports their patent portfolio and maximizes the value of their research. It turns their fabless strategy into an advantage that is harder to steal. smart.
It would not be surprising to learn that Samsung made some concession to win the 14nm business, such as promoting it in a next gen phone. Samsung is very familiar with partnering with their competitors--- see their cozy relationship with mortal enemy apple.
I'm thinking this is a huge deal for QCOM. Or at least a huge deal for their stock.
Some drink at the fountain of knowledge. Others just gargle.
They're "horrible" but they are, sadly, the best now that TI has exited the business.
MTK is notorious for giving their customers C&D letters when they dare to comply with the GPL (Google is cracking down on this with Android One, but I know of at least one non-One device that had its kernel sources C&Ded by MTK.)
Rockchip and company are no better
Samsung publishes no reference source that matches any production devices (I speak from experience here - back in 2012/2013 I was one of the CyanogenMod co-maintainers for Samsung Exynos4 devices. Every member of the team got sick of dealing with Samsung's crap and lack of documentation, we all switched to Qualcomm)
Nvidia was horrible but have improved a lot with the SHIELD family of devices, although I dislike their approach to AOSP support. They have a lot of closed-source binary HALs (just like Samsung) but at least don't hack the interfaces of those HALs in ways that break compatibility with AOSP. Unfortunately this means that if you find an issue with the HAL (such as not supporting AC3 passthrough) there's nothing you can do about it.
Qualcomm is no angel (see the Nexus 4 and Nexus 7 factory image messes), but with their CAF reference sources, they're better than anyone else currently in the business about software support.
retrorocket.o not found, launch anyway?
can anyone explain why Samsung went to the trouble of designing Exynos if they weren't going to use it everywhere?
From what I can tell they were used in their foreign mobile offerings (international Galaxy versions), possibly because of different LTE patents required for US use, and possibly due to less competition from handset manufacturers meant they didn't need to differentiate on performance. Still, I'm surprised one would go to the trouble designing their own SOC and not go to the trouble of using that R&D investment everywhere.