Samsung Developing Low-Cost 14nm and High-Performance 10nm Processes

Samsung has revealed its foundry’s roadmap for the next few years, which includes the development of a low-cost 14nm FinFET process (14LPC), with hopes that it can pass on those savings to its customers, and a new 10nm process (10LPP) that increases performance by up to 10% over its previous 10LPE process.

Samsung foundry’s process technology roadmap for the coming years includes:

  • 28nm: addition of RF and eNVM technology to our 28FDS baseline.  RF will be available this year and eNVM will be rolled out in phases in 2017 and 2018;
  • 14nm: 3rd generation 14LPC offering which provides a lower cost option, without design rule changes or performance sacrifice.  To enable connectivity features, we are also introducing RF add-on to 14LPC this year.  We have also expanded solutions on our 14nm FinFET to cover product applications in the Networking/Server and Automotive segments;
  • 10nm:  2nd generation 10nm with higher performance over 10LPE will be introduced.  We call this 10LPP.  10LPP will come with 10% speed boost, maintaining design results with 10LPE;
  • 7nm: we have already begun work on our cost optimized 7LPP node which comes with very competitive PPA scaling;
  • 8” matured node: keeping in mind there are still ample of new designs and applications that can take advantage of 8in technology, we are opening up our differentiated 8in technologies ranging from 180nm to 65nm, covering eFlash, Power devices, Image sensors and High voltage processes.

“There are always concerns about trading off cost versus performance,” Kelvin Low, Samsung’s Senior Director of Foundry Marketing, told EE Times. “LPC has the same PDK of [14nm] LPP. The number of steps has been reduced […]That allows us to achieve a lower cost point on manufacturing and we decide to share that with our customers.”

“We think 10nm will be a much longer node than other foundries are claiming it will be. We think 7nm has to be defined and optimized to be cost effective to the masses, not just the high margin products,” he added. “EUV is an important enabler for a 7m cost affordable node.”

TSMC Expects 7nm Mass Production by Early 2018

Now that 14nm/16nm chips are finally upon us en masse, some are already looking past that to the next node. Logically, the next node would be 10nm but that won’t be true for GPUs. Both AMD and Nvidia have tended to skips nodes and jump on every other node to save cost. This means today’s announcement from TSMC on 7nm holds special importance for the PC GPU market.

According to the latest shareholder report, TSMC is planning to bring forward 7nm production from previous roadmaps. This is reportedly due to a desire to beat competitors Intel and Samsung/Global Foundries to the new node. TSMC already has 20 customers lined up for 7nm, with 15 tapeouts expected in 2017 and mass production in 2018. 2 of those customers are pretty much guaranteed to be Apple and Nvidia.

Unlike 10nm which is mobile oriented, 7nm will target high-performance parts like GPUs as well. This means if TSMC hits 7nm before Samsung/Global Foundries, Nvidia has a chance to beat AMD to the new node and hold a process advantage. 7nm is expected to be 60% denser and 30-40% more efficient than 10nm so it’s a good deal better than 14/16nm. Given the difficulties Intel has faced with 10nm and how close we are to the end of silicon, it remains to see if TSMC can reach its goal.

Samsung Starts 10nm-Class DDR4 Mass Production

Much of the focus on process technology has been on GPUs and CPUs. However, there are many other products where semiconductors are needed, among them DRAM. Samsung has been and continues to be one of the leaders in DRAM production and have moved on the next process node with DDR4. According to Samsung, they have started production of DRR4 using the latest “10nm Class” process.

Samsung isn’t always the most forthcoming about their process technology so all 10nm class means is that it is below 20nm. In addition to the reduced size and increased density, the new DRAM modules also feature 8Gb capacity and run at DDR4-3200 speeds. They are also 10-20% more power efficient than current 20nm DDR4 modules. While ram power consumption isn’t a large sahre, every bit helps when you’re talking about servers and mobile devices.

Using a 300mm wafer, the new DRAM is 30% more effiecent which means DDR4 prices should fall once Samsung ramps up their 10nm class production. Samsung has also changed their DRAM design so higher speeds should be easier to come by in the future. DDR4 has just really gotten going and it looks like we’ll be seeing plenty of advancement in the years to come.

Intel Ditches ‘Tick-Tock’ for ‘Process-Architecture-Optimization’

Adopted in the troubled days of 2006 and Netburst, ‘Tick-Tock’ has served Intel well, pushing the processor firm ever forwards. Though it has served Intel well in the past, recent years have shown that the strategy has become untenable. It comes as no surprise now, that Intel has revealed the successor to ‘Tick Tock’, PAO, or better known as ‘Process-Architecture-Optimization’.

Unlike ‘Tick-Tock’, ‘Process-Architecture-Optimization’ carries a much more unwieldy title but it is the natural evolutionary step. While the old way was to introduce a new process on the Tick year  followed by a new architecture in the Tock year, PAO builds on the same idea. Under PAO, the first year is a new process node, which is based on a known architecture. This will then be followed the next year by a new architecture and finally, in the third year, Intel will optimize both the process and architecture for the final release on that process.

The perfect example is this upcoming generation where we saw Broadwell on the new 14nm node in 2014. That was followed the next year with the new Skylake architecture  in 2015 and then Kaby Lake, which is an optimized Skylake architecture still on 14nm will drop this year. The delays and costs Intel faced with 14nm and 10nm simply precluded them moving onto a new node in the 2 years ‘Tick Tock’ requires.

Moving to a 3-year cadence makes much more sense as Intel has already been slipping into a 2.5 year ‘Tick Tock’ cycle since the launch of Haswell/Haswell-Refresh. By changing it to 3 years, Intel is just accepting the reality that moving to newer process nodes is just going to be harder and longer.  PAO is the new normal as Moore’s law is dead and there is nothing anyone can do about until we move past silicon.

Intel Admits End to Scaling and Moore’s Law

Even though everyone has pretty much already seen the writing on the wall by now, Intel has remained staunch publicly. At long last, the chip giant is admitting that scaling will have to rely on improvements other than clock speeds. In fact, Intel is going as far to say that the future of semiconductors will rely on technologies that reduce power consumption rather than performance.

According to William Holt, the head of Intel’s Technology and Manufacturing Group, the semiconductor industry will see “major transitions” and the new technology will be “fundamentally different.” In order to continue moving forward, most of the new technologies mostly present a reduction in power consumption but at the cost of clock speeds due to lower switching speeds. This means all performance gains will have to come from improved IPC, new instruction sets and more cores.

With industry leader Intel already having delayed both 14nm and 10nm, it looks like silicon is nearing the end of the road. Even with the use of problematic EUV, the move to alternatives like lead telluride, carbon, Indium antimonide and indium gallium arsenide will likely happen within the decade. Even without major performance gains though, there is going to be a lot of excitement as laptops and mobile devices get better and better battery life.

Intel Cannonlake Delayed Till Mid 2017 Could Help AMD in a Big Way

Despite all the bad news from AMD thus far, it looks like they’ll be getting a break for next year. According to a rumour, Intel is delaying their Cannonlake CPUs from late 2017/early 2017 to at least the second half of 2017. Given AMD’s launch of their new Zen CPUs will take place around the original Cannonlake launch date, AMD has a chance to make their 14nm offerings before Intel launches the 10nm Cannonlake.

After Sky Lake which launched a few months ago, Intel is expected to follow up with a 14nm refresh named Kaby Lake refresh in early/mid-2016. That was expected to be followed up by the 10nm refresh Cannonlake in late 2016/early 2017. With Cannonlake now pushed back, it pretty much confirms that Intel’s Tick-Tock will have to move to at least 1.5 years and likely closer to 2. Kaby Lake is set to target sales up to week 9/18 of 2017 meaning Cannonlake will come sometime afterwards.

With 10nm pushed back, AMD has the opportunity to launch their own 14nm Zen CPUs in late 2016 against Intel’s 14nm Kaby Lake. While the process nodes are different, on marketing, this is a win and likely a better showing in terms of power and die efficiency. It also means that AMD can target the holiday season relatively unaffected by a new Intel lineup. The delay for Cannonlake also means the Sky Lake architecture successor will be pushed back as well, giving AMD more time to catch up with Zen as Kaby Lake and Cannonlake are not expected to improve IPC too much.

A delay to Cannonlake also means that the rumoured core count increase for the mainstream platform won’t be out yet, giving AMD an opportunity to compete on core counts as well if their IPC falls short of Kaby Lake. Overall though, it drives home that fact that even with their massive resources, Intel is still hitting a wall with newer silicon processes and it’s time to start looking for a successor.

Thank you Benchlife for providing us with this information

iPhone 7 to Feature Innovative new Intel LTE Chip

A new report suggests that the new Apple handset, the iPhone 7, is set to feature the Intel 7360LTD modem chip, which doesn’t sound terribly exciting, but given that it’s also reported that Intel already has 1,000 employees working on the brand new hardware for the upcoming handset, it should be pretty special.

The new LTE chip is set to give the next-gen Apple handset a hefty speed boost, offering much faster wireless speeds that I’m sure any user of the device will welcome. The new chip is said to be capable of up to 450Mb/s downlink, will support an impressive 29 LTD bands, meaning that using the phone virtually anywhere in the world shouldn’t be an issue, with all major 4G LTE bands supported.

“Sources with knowledge of the situation say that Apple eventually would like to create a system-on-a-chip (SOC) that includes both the phone’s Ax processor and the LTE modem chip. A system-on-a-chip design could deliver significant returns in improved speed and better power management.” said Venturebeat

If this works out, a partnership between Intel and Apple could be very lucrative and could open up a lot of possibilities for both companies in the future. Of course, Apple is sticking with the LTE chip for now, as the iPhone 7 is expected the run on the new 10nm A10 chip.

Intel May Increase Mainstream Core Count for Cannonlake

Intel has long limited the mainstream platform to 4 physical cores, with 8 threads due only to Hyper-Threading. This has held true for the longest time, with power users who wanted more cores making the jump to LGA 1366 and 2011. According to an Intel CPU engineer profile, Intel will start offering more cores starting with Cannonlake in 2017. Instead of a jump to 6 cores, the alleged Cannonlake SoC will feature up to 8 cores.

While some may say that this is a server processor, that is unlikely due to the SoC designation. That usually points to a mobile environment, with an attached GPU and other dedicated hardware on the same chip. The biggest question is whether or not this core-count increase is going towards mobile devices or the higher performance desktop platform. As we all know, the mobile world ahs become obsessed with core counts so a low power extreme multi-core CPU may be there for marketing. That segment would be better served though by a 10nm shrink of Goldmont from the Atom lineup which should still exist in 2017.

One of the reasons Intel may be moving to more cores on the mainstream platform in 2017, is that software is slowly starting to feed 4 cores quite well. Even with Hyper-Threading, 4 cores may simply not cut it. By pushing more cores to mainstream users, Intel is also pushing software developers who have generally been loath to make their software more-multithreaded. Another reason is that by 10nm, Intel will have enough space and cost savings to allow for 6 and even 8 cores to be cheap enough to produce. 10nm will be 4 times smaller than the 32nm process, allowing for many more chips per wafer.

Finally, 2017 is also the year that AMD will launch their Zen architecture. Set  to provide a 40% IPC over current Excavator chips, the new lineup is set to arrive on a 1x nm process and feature up to 8/16 cores +  SMT (AMD’s Hyper-Threading) on the mainstream platform.It may be that Intel perceives Zen may be a threat to their mainstream platform and is taking proper precautions. Whatever the reason, 2017 looks to be a really interesting year for CPUs, with major product introductions from both camps.

Thank you The Motley Fool for providing us with this information

TSMC Ships 16nm FinFET Chips – 10nm and 7nm On Their Way

In a move that is sure to please AMD and Nvidia fans, TSMC has started volume production of 16nm FinFET chips. According to TSMC president and co-CEO Mark Liu, the ramp up for 16nm will be more aggressive than their 20nm process, leading to improved market share for TSMC. Both AMD and Nvidia rely heavily on TSMC to deliver their chips, with the mobile focus of 20nm leading to an extended cycle on the 28nm for PC GPUs.

With the reveal that AMD has taped out their first FinFET chip, its looks like the chip was for TSMC 16nm FinFET, not Global Foundries/Samsung 14nm FinFET. With shipments for 16nm already started, AMD and Nvidia may have new GPUs set for Q2/Q3 2016, offering improved power efficiency and lower power consumption.

TSMC also revealed that their 10nm process, based heavily on 16nm, will also begin production in 2016. The 10nm process will improve frequencies by 15% and power consumption by 35%, which is understandable given that FinFETs tend to reduce clock speeds when they are introduced. 7nm is also planned for early 2018, which is a pretty aggressive schedule for TSMC.

With TSMC back on track, the future for GPUs looks brighter. Given the differing standards for semiconductors, Intel 14/10nm and Samsung 14nm are not comparable with TCMC’s. So while TSMC may reach 10nm before Intel does, Intel will actually still have the smaller process.

AMD Tapes Out First FinFET Chips – Expect Arrival Q2/Q3 2016

While much of the focus from yesterday’s financial call was the poor state of finances, another small tidbit about future plans was mentioned. CEO Lisa Su revealed that AMD has taped out their first FinFET chips back in June this year. While the exact process node has not been revealed, we can assume it is either TSMC 16nm or Samsung/Global Foundries 14nm.

From when the chip first gets taped out to initial production, it takes about a year for the chip to get to market. It takes about 3 months for the chip to get produced, and about 1 month to implement all the fixes. Then the cycle happens once more, taking another 4 months, then finally, production is ramped up, taking a grand total of about 12 months. This means we can expect the FinFET chips to arrive about June/July in 2016, assuming AMD doesn’t hit any major obstacles.

FinFETs will help AMD claw back power efficiency from Intel, who already moved to FinFETs with their mainstream 22nm process back in 2013. Given the timeframe, the chip is likely either  a GCN Arctic Islands chip or a Summit Ridge Zen based processor. Both, but especially on the CPU side, can do with better efficiency as Nvidia and Intel are ahead on that front. FinFETs can also reduce overclocking headroom so AMD will have to focus on improving instruction efficiency to compensate. In all, 2016 looks to be a good year for AMD if they can make it there.

Intel Confirms Death of Tick Tock – Kaby Lake to Replace Cannon Lake

In an ominous sign for the future, Intel has officially confirmed the delay of their 10nm process, used for Cannon Lake, till the second half of 2017. In its stead, the 14nm Kaby Lake will take its place for 2016. This confirms the earlier reports that Intel was facing difficulties in 10nm and was adding Kaby Lake to give them more time to work on Cannon Lake. This does seem to place doubts onto earlier information about “Ice Lake“.

CEO Brian Krzanich noted that the firm had already encountered difficulties with their transition to 14nm, which led to the stop gap Haswell Refresh. Kaby Lake will take a similar role as Haswell Refresh but for the Skylake microarchitecture, but will receive some “key performance enhancements” over Skylake. Kzranich also noted that the time frame for Tick Tock was likely to shift to 2.5 years instead of the previous 2 years as die shrinks get harder and harder to implement.

Intel has long kept ahead of the curve with the Tick Tock strategy, but it seems that they can no longer rely just on process technology to keep them ahead of competitors. This is the first official confirm we have for Kaby Lake, but given that the previously leaked slides have shown no real changes for the desktop SKUs compared to Skylake, these “key performance enhancements” may be limited to mobile.

Skylake is looking to be a good buy for those in the market today as it now seems guaranteed that it’s true successor won’t come till 2017. However, Cannon Lake will still be based on Skylake so really, the next major update won’t be till 2018. This should allow AMD to make a splash with their new Zen architecture in 2016 as long as it lives up to its hype. It will be interesting to see how everything plays out over the next few years.

Cannon Lake Goes Missing – Rumors Point to Ice Lake After Kaby Lake

The latest report on Intel’s roadmap shows an interesting turn of events. Intel is supposedly scrapping Cannon Lake, the 10nm chip series meant for 2017, in favour of Ice Lake. This chip will launch after Kaby Lake which itself, is set for 2016 as a replacement for the delayed 10nm architecture previously known as Cannon Lake. More interesting even is the suggestion that Intel will reintroduce FIVR, also known as Fully Integrated Voltage Regulator.

Intel first debuted the FIVR with their Haswell architecture, continuing to use it for the follow-up Broadwell. FIVR allow motherboards to be cheaper and simpler by reducing the number of different voltages that need to be supplied to the CPU package. The FVIR also allows Intel to better control voltages and in some cases, can improve overclocking. Skylake however, set to launch later this year, won’t have FVIR, opting to return to old systems. This change is due to the fact that the Skylake design team is different than the Haswell/Broadwell one.

If Intel does reintroduce the FIVR with Ice Lake, it being the direct successor to Skylake, it points to more changes than what a regular tock or die shrink would get. A return to FIVR would make some sense if we recall Kaby Lake. Kaby Lake is set to be a 14nm refresh of Skylake, meaning Ice Lake might actually be a combined Tick/Tock, meaning the Haswell team might also be working on Ice Lake, which given their past design practices, this makes sense.

At this point, this report hasn’t been confirmed yet so it might not be best to read too much into it. The constantly changing rumours coming out does seem to point to turmoil for Intel’s roadmap ahead. If Intel does slip up, it may be all that is needed for AMD to make a comeback with Zen.

Thank you HardwareLUXX for providing us with this information

Intel Faces More Delays in 10nm Production

We brought you news earlier this week that Intel appeared to have delayed their 10nm Cannon Lake chips in favour of 14nm Kaby lake. Now more reports have emerged about the delays and yield issues Intel is facing with the transition to 10nm. Word in the industry is that Intel is further delaying their 10nm ramp up into 2016 meaning it’s very likely there will be no 10nm chips till late 2016 at the earliest.

The biggest issue appears to be 10nm yields are below expectations. With low yields, the new process isn’t able to offer any new value for Intel, meaning it makes financial sense to hold off on the transition. This delay is supported by news that a planned $6 billion upgrade to Fab 28 is also being delayed, pointing to a lack of urgency to hit 10nm from Intel. If 10nm delays pile up, the death of Moore’s law and Intel’s Tick Tock strategy are just around the corner.

Intel faces stiff competition from fellow semiconductor manufacturers TSMC and Samsung, both of whom are pushing their own 10nm process. If either of those firms is able to hit 10nm first, they could have a marketing coup as Intel has long been held to be the leader in process technology. Intel would still liklely have a qualitative edge as their process tends to be more advanced at the same node and not all nanometers are equivalent.

A delay to 10nm would also hurt Intel’s mobile plans. While Intel has good mobile architecutres, the backbone is based on a superior process technology that allows Intel to outperform competitors while being more cost effective and power efficient at the same time. If ARM competitors are able to catch up with their own 10nm chips, Intel will be forced to abandon their strategy of offering expensive subsidies to sell chips. AMD also has a chance to strike back in the desktop and server space if Intel is forced to stand still. While competitors will likely hit the same wall Intel has, parity may be enough for them to catch up.

Thank you SemiWiki for the information 

Intel to Release 10nm CPUs in 2017

In the ever “shrinking” world of processors, Intel’s GM of its Middle East and North African region, Taha Khalifa, has stated that Intel is expecting the launch of these new processors to come in early 2017.

Khalifa had this to say: “We have been consistently pursuing Moore’s Law and this has been the core of our innovation for the last 40 years. The 10nm chips are expected to be launched early 2017”.

For those looking for an upgrade sooner, expect Intel to debut the new 14nm Skylake processors later this year. While there had been murmurings of the 14nm processors being launched in 2013, various technical setbacks with the architecture used in Broadwell delayed it till 2014. This delay affected many of Intel’s forthcoming processors, including the 10nm Skylake family. Coming first under the 10nm umbrella will be tablet and mobile processors, followed by the desktop family later in 2017.

Thank you TweakTown for providing us with this information.

TSMC To Boost 20nm Production During Rest of 2014 – Good News For AMD/Nvidia

There’s good news for AMD, Nvidia and graphics card enthusiasts because the Taiwan Semiconductor Manufacturing Company (TSMC) is announcing increased capacity for the second half of this year. TSMC claim 20nm production will ramp up in Q3 to account for 10% of their revenues and by Q4 that will be 20%. TSMC are also looking to ride the global semiconductor wave which is expected to rise 3-5% annually over the next half decade, TSMC’s growth rates will be higher than the industry average.

In 2013 TSMC’s production reached 15.67 million 8 inch equivalent wafers, up from 14.04 million in 2012. TSMC is currently pushing towards 20nm production, is starting early roll-out of 16nm production and is researching experimental 10nm production which it hopes to mass produce by 2016 and have trial production of in 2015. The future certainly looks bright for TSMC and if you’re a graphics card enthusiast this is only good news – smaller process nodes means the potential for more performance. Bring on 20nm I say!

Source: Digitimes

Image courtesy of ExtremeTech.com