Common Platform Technology Summit: mobile, wireless and the path to 20nm
For those of you who may not be familiar with the Common Platform; it is an alliance formed by IBM, Samsung, and Global Foundries to collaborate on development and manufacturing of advanced CMOS processes. Infineon, Freescale, STMicroelectronics and Toshiba also participate in the related Joint Development Alliance. The three Common Platform partners, plus STMicro, are working to “synchronize” their foundries for 28nm CMOS so that a design could, theoretically at least, be manufactured at multiple fabrication facilities around the world. I say theoretically because mask cost is no small consideration at 28nm, and there is no assurance of mask portability from foundry to foundry. Compatibility, in this case, means portability of the GDS-II layout database.
On Tuesday January 18th, the Common Platform partners held a Technology Summit at the Santa Clara Convention Center. The morning agenda consisted of a series of keynote addresses from the three partner companies, along with presentations from ARM and Qualcomm. While Qualcomm ’s presence should not be surprising, since they are the largest fabless semiconductor company in the world, their participation is also indicative of the repeated emphasis on mobile, wireless applications that took place during the conference. There is no doubt that cell phones and other portable connected devices have taken over as the #1 driver of the semiconductor industry.
Dr. Stephen Woo, President of the System LSI Division at Samsung
Dr. Woo shared data from industry analyst forecasts showing that 55% of all mobile handsets will be smartphones by 2015. As has been widely covered in the wireless industry press - feature phones will be no more. Woo also predicted that tablet computers will soon replace notebooks and netbooks. This shift in computing platforms to mobile devices with smaller form factors will lead to a reduction in global energy consumption, according to Dr. Woo.
The emphasis on mobile device and low power requirements was used to highlight the capabilities of Common Platform’s low power 28nm process. This process is a direct shrink from 32nm, using HK/MG (high-K/metal gate), which was said to reduce leakage by 10X. As an example of the performance which can be achieved, an ARM Cortex A9 test chip was able to run at 1.6GHz with 1 volt Vdd at the 32nm process node. (As a comparison, the most powerful Android smartphones that were recently announced at the Consumer Electronics Show, such as the Motorola Droid Bionic, will use a 1GHz dual-core Cortex A9 processor that is fabricated in a 40nm process).
Chia Song Hwee, Chief Operating Officer, GLOBALFOUNDRIES
Mr. Hwee continued the mobile application focus, stating that the 32nm ARM Cortex A9 has been validated, and that a 28nm test chip is in progress. The28nm shrink will double density, and provide up to a 50% increase in speed along with lower power. Other test chips that are in process include AMD’s Liano quad core CPU with a Direct-X GPU, which is expected to hit volume ramp this year.
Dr. Gary Patton, Vice President, IBM Semiconductor Research & Development Center
Dr. Patton spoke on the “Technology Innovation at 20nm and Beyond“. The challenges that he listed for moving from the 28n node to Moore’s next stop at 20nm include:
- The end of scaling
- Escalating cost
- Requirements for material innovation
The escalating cost of developing each new process node is the prime motivation that bring the participants together in the Common Platform coalition.As Dr. Woo said in his presentation; there will be a total of 7 companies at the 32/28nm process nodes, but he predicts that only 4 companies will be left with at 15nm.
While listing a number of candidates for new material innovation; such as silicon nanowires, copper plating, and monolayer thick gate stacks, most interesting in my mind is the use of “computational scaling“. Whereas previous process nodes advanced based on photo-lithographic innovations such as wet immersion and interference patterning, 20nm will rely on “source mask optimization“(SMO).
Optical Proximity Correction, or OPC, has been in use since the 130nm node. OPC algorithmically alters a mask’s physical features to compensate for limited resolution of the light sources used in wafer fabrication. With SMO patterning is controlled pixel by pixel, using multimirror array process that IBM developed through a 2-year long project with ASML, Zeiss, and Mentor Graphics.
Dr. Patton laid out this timeline for extending Moore’s Law further:
- 2010 - 28nm
- 2012 - 20nm with SMO
- 2014 - 14nm with double exposure and a change to EUV.
- 2016 - 11nm, with EUV.
The most controversial change for Common Platform’s 20nm development is a switch to a “gate last” fabrication sequence for the 20nm node. All process nodes through 28nm have been “gate first”, and Dr. Patton laid out the reasons why that was the right choice in his presentation. He justified the change by stating that “20nm demands different requirements than 28nm“, and then broke down the issues that were weighed in making the decision:
- Density: 20nm LP patterning is dominated by RDR (restricted design rules).
- 10-20% of the benefit of gate 1st is lost at 28nm.
- Process simplicity would be better with gate 1st.
- Power/performance - gate 1st helps (no strain, low cost).
- For Low Power, strain engineering, and local interconnect, gate last helps.
The bottom line.. it’s all about achieving lowest possible power. That is the most important requirement for mobile devices, even if cost and performance have to suffer in the… umm… process.
Mike Muller, Chief Technology Officer, ARM
Mr. Muller’s presentation was titled “From millimeters to kilometers“, an interesting vision of the future of ubiquitous computing.He included an amusing look back at a true visionary, H. Osborne of AT&T, who made a number of uncanny predictions way back in 1956. It’s amazing to see how many have come true:
- Everyone will be given a telephone number at birth. While not exactly a reality, how many children now carry cell phones in elementary school?
- Musical ringtones.
- All electronic switching.
- Robot dialing (i.e. voice activation?).
- Worldwide connectivity.
- Video phones.
In “millimeters to kilometers“, Mr. Muller spoke of the next wave of computing being the “internet of things”, with vast sensors networks - including embedded in the human body. This vision is one of KHz and MHz processors powered by energy scavengers. Ubiquitous computing environments could be built with paintable computers.
Mr. Muller provided an example in answer to his own question of “What kind of sensors can you build“? He showed a 1mm3 interocular pressure sensor, consisting of a 3D stack with 2 solar cells, a Li-ion battery, and an extremely low power ARM Cortex-M3 operating “near threshold”. The implantable device integrates the solar cell, antenna, CPU and digital logic with a battery and pressure sensor to perform data logging of pressure changes that provides valuable insight into the condition of glaucoma patients as they go about their lives under real world conditions.
In another example of a wireless sensor network; a 65nm Cortex M0 was integrated with a UWB radio and 10K of 180um retention RAM to provide an earthquake logger that could be attached to buildings.
In Mr. Muller’s view, ubiquitous environments arise from the development of computing storage in the cloud, combined with services. It is the services, he said, that will enable behavioral change. Hardware is the easy part, he said, software is the problem.
David Steel, Executive Vice President of Strategic Planning & Corporate Communications, Samsung Electronics America
Picking up on themes presented by Samsung at CES, Mr. Steel talked about “Smart + Connected“: “Smart TVs” and smartphones. Super phones (a term coined by Google with the Nexus-1 and now used by NVIDIA and others) will be the “connective tissue” to communicate with other devices and services, he said. His prediction was that 39% (love these forecasts… why not 40%?) of cell phones worldwide in 2014 would be smartphones, but ¾ of U.S. market in 2014
In his view the “changing computing paradigm” consists of vertical markets (such as medical, education, enterprise), with more connectivity options, including subscription, WiFi, and session-based 3G/4G.
Other highlights of Mr. Steel’s talk, some of which can be considered wishful thinking for rapid adoption of connected TVs:
- Smart TVs changing the video value chain, IP content,TV as the hub of home entertainment. (This would be a return to the living room, given data on the shift to PC-centric entertainment and the internet).According to Mr. Muller, 50% of consumers with “Smart-TVs” are now using them connected to the internet.
- 4G adoption increasing. From near-zero before, that would appear to be a given. However, with Samsung providing phones to all the major carriers, and with AT&T and T-Mobile now relabeling their 3G networks as 4G, Mr. Steel made a point to include both HSPA+ and LTE in his definition of 4G. He forecast an LTE “inflection point before 2014” of 20M users.
- Device-to-device (or M2M) is growing, but there is a shift to cloud and web-centric services & applications.
- Smarter devices means more context-aware. “Sensors, sensors, sensors” drive new use cases.
Dr. Jim Thompson, Qualcomm, Sr VP Engineering
Wrapping up the morning’s theme (which made the Common Platform Summit look more like a CTIA Wireless show than a semiconductor fabrication technology conference), Qualcomm’s Dr. Thompson spoke of “expanding smartphones to the masses“, from <15% of subscribers in 2009 to >45% in 2014. reinforcing the theme of the morning, that mobile devices now drive the semiconductor industry, he pointed out that more than 45% of all silicon content is now consumed by wireless handsets. With this has come a shift in application development away from the PC to smartphone platforms.
Dr. Thomson spoke of Qualcomm developing a multi-mode chip for LTE, saying that the company is “uniquely positioned“ to deliver the 1st multimode capable of supporting 4G with LTE, along with WCDMA & CDMA2000 networks. This is an interesting development, given rumors of Qualcomm providing a multimode chip for the next-generation iPhone.
Dr. Thompson also addressed the benefits of integration of a baseband and applications processor on a single chip, poining to a smaller footprint, and lower cost for the chip as well as the complete system, saving $3-$6 in this price-sensitive market. The Qualcomm MSM8960, arriving about 18months after its original taarget date, provides a multi-mode modem, and is the company’s 1st 28nm device. Performance is claimed to 5X previous generations in CPU speed, with 75% lower power, 4X the graphics perfromance, as well as LTE, WiFi, GPS, BT, and FM radios built-in.
Currently no items