Semiconductor industry executives converge to discuss a potpourri of problems—and potential solutions—that will have an impact on future systems.
With a theme of “The New Frontier” for the Semico Summit 2011, a pervasive response from presenters was change and coping with change. However, recurring themes such as “smart,” partnerships, and 3D packaging made their presence felt, too. Above all, what shone through is that despite today’s underwhelming economic landscape, stagnancy hasn’t befallen technology.
As Jim Feldhan, president and CEO of Semico Research Corp., noted when he kicked off the summit, “Now that we are all connected, we want to be even more connected.”
With five cell phones sold for every PC, Feldhan can’t help but be excited about cell-phone technologies. A cell phone is replaced every 18 months, and newer smartphones are fueling an explosion in content demands. The smartest phones provide microelectromechanical systems (MEMS) opportunities for a host of devices: gyroscopes, accelerometers, auto-focus actuators, pressure sensors, micro mirrors, silicon microphones, film bulk acoustic resonators (FBARs), RF switches, digital compasses, oscillators, micro speakers, temperature sensors, and others.
Feldhan sees total MEMS and sensors growing at 58% in 2011, with products such as gyros integrated with accelerometers, and/or pressure sensors combined with digital compasses. His bottom line is semiconductor market growth of 8% in 2011 and 2% in 2012.
Transitions—FPGAs, Packaging, Connectivity
Moshe Gavrielov, president and CEO of Xilinx Inc., identifies the insatiable need for bandwidth, ubiquitous connected computing, and the programmable imperative as key market drivers. He notes that by 2015, two-thirds of mobile traffic will be video, triggering a 92% CAGR in mobile data traffic. Specifically, machine-to-machine traffic, which conveys wireless data from smart meters, security cameras, health care, telematics, and other new sources, is expected to grow at a rate of 109%.
The impact on field-programmable gate arrays (FPGAs) is critical to Gavrielov. He notes that an FPGA crossover product can be developed in 11 months, versus 20 months for an application-specific integrated circuit (ASIC) with less than half the resources. The FPGA crossover will bridge into processor, analog/analog mixed-signal, application-specific standard product (ASSP), and ASIC territories.
Key enablers for integration of crossovers are monolithic multicore CPU and programmable-logic as well as 3D packaging systems. As part of a comparison to history, Gavrielov noted that the “plug and play” of the mid ’90s didn’t materialize then, but it is happening now with FPGAs. “In a rapidly moving world, you need to make sure that your product definitions are fluid, so you can change them on an ongoing basis,” he says.
Joe Sawicki, vice president and general manager, Design-to-Silicon Division of Mentor Graphics, observes that cost reductions from device scaling have enabled pervasive computing (Table 1). Today, scaling has moved into the third dimension. One technique for this type of scaling is a packaging interposer (Fig. 1).
A packaging interposer poses problems for unique routing and verification requirements, though. Nonetheless, Sawicki sees a rapid rise in the use of interposers within the next two to three years for high-end computing technology, as well as mixed analog, RF, logic, and memory in multi-die stacks.
Sawicki left the audience with some changes that could happen in the future (2026) based on the continued scaling ability of semiconductors. Those projected changes raise some interesting questions:
• With 500 MCUs in a car, are you still driving it?
• With 1-Gbit transmission speed, will you be able to deliver a movie to the home in a blink of an eye?
• With a movie encoded in 4.7 minutes, what device won’t have video?
• With a function’s cost sitting at 0.015 cents, will that drive single-use electronics?
Sawicki also added a caveat: “It’s going to take more than scaling to achieve it.”
Len Perham, president and CEO of MoSys, says that with the Internet increasing fourfold from 2010 to 2013 and exceeding 50 exabytes (1 billion gigabytes) per month, memory becomes a major system bottleneck. His solution involves the end of parallelism and a transition to serial data flow. He offered the example of MoSys’ bandwidth engine IC, which enables serial chip-to-chip communication at the board level and improves packet efficiency. It also reduces access time via multi-partition, multi-bank architecture.
Danny Biran, senior vice president of Altera, observes that the lines between product categories were clearly defined in the past, but now those lines are blurred, creating “converged platforms.” As a result, semiconductor companies will only succeed by creating new types of products. One example is a semiconductor product that includes advanced processor technologies, such as the Intel Atom co-packaged with FPGAs.
Converged platforms with processor and memory cores and FPGAs represent a new class of solution. “Those customers that are willing to change the way they are organized and the way they define their architecture, the way they implement their systems, are doing much better now than their competitors who are still falling into the trap of ‘this is how we always did it,’” says Biran.
Connectivity is an issue from many perspectives. “The internet is becoming the plumbing of our planet,” says Bob Krysiak, executive vice president and general manager for the Americas Region of STMicroelectronics. He notes that mobile payment transactions will grow 68% from 2010 to 2015.
Healthcare is an important area for many electronic companies, because spending in this area is 8% to 16% of developed countries’ GDPs. Of that, 75% to 85% involves chronic disease management. For example, there are over 300 million diabetics in the world. Thus, patient monitoring, including connectivity to a wireless gateway, has become big business (Fig. 2). Forecasts show that there will be over 50 billion connected devices by 2020; as such, Krysiak expects security to play an increasingly important role.
Gregg Bartlett, senior vice president of Technology and Research and Development at GLOBALFOUNDRIES, notes that with the promises of social networking and cloud computing, “silicon plays a pivotal role as it always has.” He adds that we must have “gigahertz capability without burning your hands up.”
Citing similar statistics to other presenters, Bartlett says that by 2014, there will be approximately 3 billion 3G subscribers. This will require solutions to address the issues associated with connectivity growth, an advanced process fab that costs from $5 to $7 billion, and customer chip design that costs from $40 to $50 million without considering $50 million for software development.
With costs for libraries and intellectual property (IP) for 32-/28-nm nodes climbing to over $250 million and leading-edge process development reaching $1 to $2 billion and requiring three to four years of development, the cry is growing louder for a new approach. Bartlett says the solutions involve materials such as high-k metal gate (HKMG), lithography advances, including rigorous co-optimization of physical design and process computational scaling. In the future, package integration is not an afterthought, he insists, because “packaging can easily reach into 35% to 50% of the cost.”
Companies and experts must address these complex issues within the confines of a new collaborative development model. For the foundry business, Bartlett contends that the traditional model of a geographically centralized foundry will transition to a distributed foundry network. Instead of a regional talent pool, there must be global sourcing. Instead of pre-packaged technology, there must be optimized design solutions created from collaborative innovation instead of homegrown R&D.
Aart de Geus, chairman and CEO of Synopsys Inc., envisions a future of “smart everything” (Fig. 3). With this trend, he sees semiconductor content staying above 25% for the next few years after hovering around 21% to 23% for the years prior to 2010.
To cope with increasing complexity of semiconductor design and processing, the issues below 45/40 nm involve power, verification, software, variability, yield, and 3D packaging. As noted by several previous presenters, collaboration is crucial toward solving these issues. Since software is one half of the chip’s time-to-market, acceleration of software development via virtual prototyping and FPGA-based prototyping becomes essential.
“If we could just move that software bubble a little bit to the left \\[in the development timeframe\\], in other words, slide the software development and verification before the hardware is available, that would have a substantial impact on this overall picture,” says de Geus. And, indeed, a massive effort is now underway to do that more systematically.
MCUs, Sensors, And The Cloud
Ganesh Moorthy, chief operating officer at Microchip Technology, points out the difference between obvious computing examples of smartphones and tablet PCs, which represent the tip of the iceberg at about 800 million units per year, versus the 10 billion embedded units per year in consumer, automotive, office automation, industrial controls, and telecommunications applications. For example, in 2010, an average car featured about 40 to 45 microcontroller units (MCUs), which is about three times the average number just a decade earlier (Fig. 4).
In consumer applications, 2010 sales for MCUs was 3390 million units/year, up from 1913 million units/year a decade earlier. From garage-door openers to thermostats, MCUs can be found in some of the most basic appliances and certainly in the more complex ones. Moorthy envisions:
• More integrated features at lower cost
• Higher performance with lower power
• Smaller size
• Wired and wireless connectivity
• High-quality, low-cost graphics
• Touchbuttons, touchsliders, touchscreens
• Energy-efficient building blocks as innovation enablers for the future
This segment needn’t contend with the leading- or bleeding-edge technology issues, but must continue to integrate and reduce the cost per function.
Sandeep Vij, president and CEO of MIPS Technologies is also convinced that all devices are becoming “smart”—Internet-connected and networked together. “Anything worth having in the future is going to have an IP address,” he says. “And will be able to get networked and linked.”
In this ubiquitously connected world, customers will look for a seamless user experience between connected devices. One answer is a single platform like a smart Android OS.
However, Vij believes that smarter devices need smarter software. Due to the ever-present design challenges of increasing complexity, emerging requirements, and time-to-market and cost pressures, he concludes that there’s a greater need for third-party IP. By allowing companies like MIPS, which uses high-end technologies such as 64-bit, multithreading, and multicore technologies to create IP, licensees can focus on their special sauce and provide differentiation. The ease of the user experience is a special case of extreme differentiation.
“When it is easy to access and also tremendously useful, it gets a magic moniker of being cool,” says Vij. “Once it becomes cool, it enters into an entirely different era of sociological behavior that actually makes these ‘must have’ items.”
Tom Deitrich, senior vice president and general manager of the RF, Analog & Sensor Group at Freescale Semiconductor, notes that sensors in 2011 consumer-electronics products are quite different than those available five years ago. In the older CE device, sensor applications were simple but innovative—a single-axis MEMS transducer with a signal-conditioning ASIC provided the sensing functionality for airbag crash detection or screen orientation of a handheld CE device.
“A sensor in 2011 is a very, very different beast,” says Deitrich. “Now we are talking about a three-axis accelerometer, a three-axis gyro, a three-axis magnetic sensor, a pressure sensor, and an altimeter all put together in one device, plus on-board power management, a signal-conditioning ASIC, connectivity, and a microcontroller.” Figure 5 shows an example of Freescale’s packaging technology for a sensor system.
Cloud-based services are enabled by sensors, analog, and connectivity semiconductors, says Dietrich. Networking numerous sensors to appropriate computing capability will provide self-aware computing for new applications. Beyond the frequently cited remote health and activity monitoring and cooperative (smart) highways, he noted how sensors and wireless communications play an enabling role in augmented reality for enhanced gaming, education, business, security, medicine, and entertainment applications.
Siddharth Sheth, vice president of marketing, High-Speed Connectivity, Inphi, concluded the 2011 Summit with a look at future requirements for high-speed connectivity. Referencing the previously mentioned projection of 50 billion connected devices, he noted that as of 2009, less than 1% were connected. Thus, there’s a long way to go.
Companies need to drive innovation to close the gap between today’s traditional solution versus tomorrow’s solution. Solving signal-integrity challenges is essential, and that involves better signal detection, faster throughput, improved error margin, and higher signal integrity. Overall, Sheth envisions a number of changes (Table 2).
One key metric is that 100G Ethernet must go from today’s 20 W to less than 8 W in the future. On that front, Inphi developed a low-power serial-deserializer (SERDES) that can address 100G Ethernet.
The semiconductor executives at the Semico Summit 2011 control many of the key technologies for today’s high-tech products. To create the kind of future they envision, though, they must knock down the semiconductor hurdles standing in the way.