TweakTown NewsRefine News by Category:
I don't know where my wafer is - come on, TSMC - but seriously, the company has just shown off two wafers at Techcon: the first, a production 20nm SoC and the second, a pre-production 16nm without any qualifiers.
Our friend over at SemiAccurate snapped the beautiful shots of things to come. 16nm is the most interesting here, as we don't have 16nm tech in our devices, but 20nm-based chips are already baked into some devices on the market. SA says that 16nm FinFETs from TSMC should arrive in early 2015 if there are no glaring problems ahead.
The shift is beginning, with chipmaker leader Intel moving into manufacturing ARM-based processors next year. Intel will expand on its current relationship of allocating some of its plant capacity to contract-manufacturing, pumping out a quad-core part based on the ARM architecture.
The Stratix 10 SoC will be manufactured for Altera, and it will be based on Intel's most advanced 14nm process, and based on ARM's 64-bit Cortex-A53 platform. The chipmaker has said that this is the first chip manufactured by anyone, and is the first 14nm Tri-Gate-based processor out of their fabs, even including Haswell which is built on the 22nm process with Tri-Gate.
The new SOC will be targeted at high-end networking and communications equipment, so it's something we probably won't be seeing in our consumer devices at all. This means Intel won't be competing in the same space as NVIDIA, Qualcomm or Samsung. It is interesting to note that Intel is now pumping out ARM-based SoCs, which is an interesting move considering its Bay Trail platform is to continue spreading into more and more devices as time goes on.
Thanks to a leaked slide, we know when to hopefully expect AMD's upcoming, next-generation APU, codenamed "Kaveri." Kaveri will be officially announced on December 5, but it will see a proper launch in February 2014.
Between its announcement and its launch, we should expect the chipmaker to ship Kaveri APUs to its OEM vendors before 2013 is a wrap. The new Kaveri APUs will be slotted into the new socket FM+ package, which will support AMD's new A88X chipset. Kaveri will feature CPU cores based on the "Steamroller" architecture, which provides an approximate 10% increase in performance over the "Piledriver" architecture.
The new Kaveri-based APUs will feature a Graphics CoreNext-based GPU strapped on, TrueAudio technology, support for even faster DDR3 memory, and PCI-Express 3.0.
During IDF back in September, Intel announced its new 14nm Broadwell processor and said that we should see it begin to ship to OEMs sometime around the end of 2013. Unfortunately, that will not be the case. During the company's quarterly earnings call, Intel said that Broadwell has been delayed until the first quarter of 2014.
The delay is due to a recent issue with defect density in the silicon wafers that were being produced. Intel says that this issue has been fixed and yield levels are back up to an acceptable number. Some analyst speculated that the delay was actually due to an excess of Ivy Bridge inventory to which CEO Brian Krzanich replied "absolutely not." From the beginning, Broadwell has been expected to go retail in 2014, and a setback of three months is not that big of a deal.
Silicon Motion has just announced the launch of a new cost-effective USB 3.0 controller for use in flash drives. The new SM3267 integrates an embedded crystal-based oscillator, power ICs as well as the control silicon all into a tiny package that lowers overall system cost. This new SM3267 also enables industry-leading data transfer rates with speeds up to 160MB/s read, a 50-percent improvement over most other single-channel controllers on the market today.
"We are very excited to introduce SM3267, our first crystal-less USB 3.0 solution with integrated power ICs," said Wallace Kou, President and CEO of Silicon Motion. "SM3267 offers superior performance with competitive cost when compared to other USB 3.0 controllers in the market, and we believe our solution will help accelerate the market in transitioning from legacy USB 2.0 flash drives to the latest generation USB 3.0 drives. This is another example of Silicon Motion leading the market in bringing new, advanced technology to the mainstream market. We are pleased to announce that SM3267 has received design-ins from most of our current USB controller customers, including many top-tier OEMs, and we expect SM3267-based USB 3.0 flash drives will be commercially available starting in the fourth quarter of 2013."
A new report has surfaced that suggest AMD is working on finalizing a consumer-focused System on Chip for mobile devices. The SoC is said to feature a combined core design comprised of ARM Cortex A57 and A53 cores and will be deployed in tablets and high-end smartphones this winter.
The combined core design should prove to be interesting as the A57 architecture is designed around a multi-issue pipeline that is capable of executing commands out of order, while the A53 is a much more docile core that runs instruction sets in order very efficiently. Graphics are said to be handled by a GCN architecture GPU, which is the same technology that powered the Radeon HD 7000 series of video cards. AMD is expected to launch the new SoC sometime in Q4, but not much else is known.
We are used to the usual teardowns, that usually involve ripping a physical device apart, but iFixit has torn down the actual A7 silicon that we find in the Apple iPhone 5S. iFixit used an Ion Beam Etcher, which takes layers off of a semiconductor, analyzing just how it is made. iFixit worked with Chipworks on the A7 teardown, discovering quite a bit:
- First, a biggy: "We have confirmed through early analysis that the device is fabricated at Samsung's Foundry. We suspect we will see Samsung's 28 nm Hi K metal Gate (HKMG) being used."
- The distance between each of the chip's transistors is 114 nm, compared to the A6's 123 nm.
- That 9 nm difference - from a 28 nm process to a 32 nm process - means the same computational power can be squeezed into just 77 per cent of the original area. But given the A7 is larger than the A6, it's clear where all that poke came from.
- In total, the A7 packs one billion plus transistors onto a 102 square millimetre field.
- Chipworks claims that the M7 section of the silicon is an NXP LPC18A1 - part of the LPC1800 series of high-performing ARM Cortex-M3 based microcontrollers.
- In fact, the M7 packs accelerometer, gyroscope, and magnetometer, and does some fancy maths before neatly passing orientation data to the main chip.
- The Wi-Fi module is exactly the same as that in the iPhone 5.
- The 4G LTE modem uses two chips: a Samsung-fabricated LTE baseband processor and a Samsung DRAM module to retain carrier specific information.
Today we got a glimpse of AMD's new 45W TDP Richland APUs thanks to a Connecticut retailer accidentally listing SKUs, specifications, and prices for two of the upcoming low-power devices. The listing detailed the new A8-6500T and A10-6700T APUs, which are both FM2 socket APUs that are clocked at 3.1GHz and 3.5GHz, respectively.
Both the A8-6500T and the A10-7600T feature a 4MB cache and a TDP of just 45 Watts. Pricing for the new chips appears to be set at $108.59 for the retail boxed version of the A8-6500T, while a systems builder tray would run customers $114.58 per APU. The A10-6700T is a bit more expensive with the tray version running $166.97, while the retail version would run $155.91. With all of the information released last week on Ivy Bridge-E from Intel, does AMD still stand a chance, or will AMD return to the throne once again with these new APUs? Let us know in the comments.
IDF 2013 - Shortly after his first stop at Intel's integrated graphics booth, Trace Hagan was able to stop by another Intel IGP demo where we were shown a Haswell processor transcoding 4K video faster than real-time speeds. The encoding process used less than 5-percent of the CPU and attributed this massive performance increase to Intel's Hardware Encode and Decode engine.
IDF 2013 - As usual, Intel gave us a glimpse at its upcoming Broadwell and 14nm process. Specifically, the chip making giant pitted Haswell versus Broadwell in Cinebench and showed that the Broadwell system consumed less power. While they didn't specifically call attention to it, you can clearly see that the Broadwell system is completing the benchmark quicker than the Haswell system, though it's not too much faster.