Expect More Cores, Less Power, And Faster Serial Links In 2011
Microspace MSMST
TILEncore board
SM10000
LPC4000
C8051T606
SATADIMM
Tower development platform
EZ430 Chronos
Forecasting is a challenging task because highlighting what’s new often means examining the past. This is especially true this year, as the playing field is changing and new players are emerging but the trends remain the same as last year.
Multicore is now a matter of numbers rather than speculation. Power is minimized while performance is maximized. And, banks of serial links remain the dominant form for high-bandwidth connectivity.
Furthermore, designers can expect a couple of other trends to continue, such as a blending of FPGAs and CPUs and the emergence of high-density, nonvolatile storage using MRAM or phase-change technology.
FPGA With Processors Or Vice Versa?
Intel’s E600C is not the first micro to be combined with an FPGA, but it’s likely to spawn more competition in this arena. It’s already showing up in platforms like Kontron’s Microspace MSMST PCIe/104 board (Fig. 1).
The Microspace PCIe/104 single-board computer (SBC) houses the Intel Atom E6x5C series processor. The E6xxC chips combine an Atom system-on-a-chip (SoC) core with an Altera FPGA. PCI Express (PCIe) links tie the two together. The stackable PCIe/104 runs the other Atom PCI Express links up the stack.
FPGAs with hard-core processors aren’t new, but typically, a host CPU was off-chip. On-chip CPUs provided programmatic support. The E6xxC puts the Atom in the primary position because of the surrounding, hard peripherals including the graphics and audio support.
The FPGA can be viewed more as an accelerator or custom I/O interface, though it’s even possible to drop in a soft-core processor for a custom multicore solution. The possibilities are intriguing.
Expect even more Arm hard-core processors inside FPGAs. Microsemi’s SmartFusion FPGA incorporates a Cortex-M3 along with a programmable analog subsystem (see “FPGA Combines Hard-Core Cortex-M3 And Analog Peripherals"). This is a significant amount of fixed logic combined with an FPGA.
Xilinx (see “Xilinx Unifies FPGA Line”) and Altera (see “Climb On Board Next-Generation FPGAs”) will be taking the 28-nm route for their FPGAs. Hard-core processors with these FPGAs will finally see the light of day this year.
Achronix has hitched its wagon to Intel and its 22-nm technology (see “Speedy 22-nm FPGA Packs PicoPIPEs”). Achronix will challenge Xilinx and Altera in the top end, but the Achronix product line does not target the low end.
The compamy to watch is Tabula. Its ABAX FPGA uses a dynamic reconfiguration approach by design where the underlying fabric can change every clock cycle (see “FPGAs Enter The Third Dimension”).
Continue on next page
Lattice Semiconductor will continue to play a major part in the FPGA competition. However, its MachXO2 PLD will be making waves in the low end of programmable devices this year (see “PLDs Get 65-nm Flash Technology Boost”). Again, hard-core logic provides a more efficient platform for developers while speeding time-to-market and lowering intellectual property (IP) costs.
More Multicore
Intel’s Sandy Bridge will give AMD’s multicore offerings a run for their money (see “IDF Highlights Sandy Bridge And Tunnel Creek”). These x86 platforms will continue to provide the mainstay for servers in the enterprise and cloud, but their core count will still be edging into the double-digit range. Multicore on the desktop, laptops, and tablets is likely to remain in the quad and hex core region as software for these platforms has yet to take effective advantage of multiple cores.
In the meantime, platforms like Tilera’s TILEncore (Fig. 2) are packing 64 cores into one chip. This year, Tilera will be delivering 100-core chips. Multicore platforms are taking on chores from deep packet inspection to weather simulations.
SeaMicro is not in the semiconductor business, but its SM10000 (Fig. 3) highlights many-core clusters in a different fashion (see “10U Rack Packs 512 Atoms”). The platform packs in 512 1.6-GHz Z530 Atom processors and 1 Tbyte of DRAM. Such a platform would be even more cost-effective with the new E6xx Atoms and potentially more power using the E6xxC Atoms with a built-in FPGA.
Dual-core and quad-core processors will be even more common in midrange and mobile computing platforms. Based on the Cortex-A9 MPCore, Nufront’s dual-core, 2-GHz NuSmart 2816 will find homes in laptops and tablets (see “2-GHz Dual-Core Cortex-A9 Targets Tablets”).
Also, a variety of quad core ARM-based platforms will be shipping in 2011 like Marvell’s Armada XP (see “ARM Microcontroller For The Cloud Delivers Quad Cores And Quad Peripherals”). These platforms will challenge the x86 dominance in servers.
NXP’s LPC4000 experiments with asymmetric multicore designs that are common in ASICs but relatively new for standard micros (see “New Platform Approaches Deliver Top Digital Designs In 2010”). The LPC4000 (Fig. 4) mixes a Cortex-M0 core with a Cortex-M4 DSP core.
Not all vendors agree that mixing cores is a good idea. However, the success of the Texas Instruments OMAP line of CPU/DSP dual-core platforms in a range of applications is a good indication of where the LPC4000 might succeed (see “Hawks And Beagles Get Board”).
GPUs And SuperComputing
Graphics processing units (GPUs) will remain graphics processors at their core. Their use as computing elements is well documented, though, especially in the area of high-performance computing (HPC) where even college students are taking advantage of them (see “What Kind Of Super Computer Can You Build With 26 Amps?”).
Software such as Kronos OpenCL is providing better support for multicore programming in general and GPUs more specifically (see “Parallel Programming Is Here To Stay”). MATLAB GPU support is part of the MathWorks Parallel Computing Toolbox. This support addresses NVidia’s GPUs such as the Tesla 10-series and 20-series (see “SIMT Architecture Delivers Double-Precision Teraflops”).
Continue on next page
Standalone GPUs deliver the performance needed for gaming and HPC, though GPUs that are part of an SoC will be more widely deployed. Typically, these GPUs handle the heavy lifting for multimedia delivery but can often be used to augment computation and communication chores. More novel software uses will crop up as developers move past the basic streaming content delivery. GPUs and SoCs mate well because less power is required to provide improved playback.
Power-Sipping Micros
Multicore may garner the glamour. But single-core, low-power micros will continue to dominate the embedded space, handling everything from wireless sensors to motor control to ZigBee-based RF4CE remote controls (see “RF4CE Revolutionizes The Remote Control”).
The plethora of 8- and 16-bit micros is not waning but definitely getting smaller in terms of package size. The crop of 2- by 2-mm microcontrollers like Silicon Laboratories’ 8051-based C8051T606 mixed-signal microcontroller (Fig. 5) that are readily available continues to grow, allowing them to fit into almost any design (see “Smallest MCU Hits 2- By 2-mm Form Factor”).
Likewise, power scavenging is on the rise, eliminating the need for batteries. Low power requirements often make a single battery sufficient for some applications that have lifetimes in decades.
The area to track this year will be the low-end 32-bit microcontrollers that are moving into this arena in force like the Arm Cortex-M0 (see “Cortex-M0 MCU Challenges 8- And 16-Bit Micros Across The Board”). Low-pin-count, small-form-factor Cortex-M0 chips will be hitting the competition where it hurts: power, performance, price, and packaging.
Storage Choices
Flash remains king in the nonvolatile storage arena for at least another year. Flash controllers like the SandForce SF-1500 SSD controller turn mutlilevel-cell (MLC) solid-state disk (SSD) drives into enterprise storage platforms (see “Thanks For The Memory”). Single-level cell (SLC) will maintain its edge but will be challenged by MLC’s price and capacity advantages.
Flash storage is leading to some interesting designs that embedded developers can take advantage of. For example, Viking Modular Solutions’ SATADIMM (Fig. 6) puts a SATA II flash drive into a DDR3 240-pin dual-inline memory module (DIMM) form factor. The DIMM actually has a SATA II connector and only draws power from the DIMM socket. The SF-2500 and SF-2600 SATADIMMs utilize a SandForce SSD processor.
Why use a DIMM slot? Many motherboards have sufficient DIMM slots to handle a processor’s memory needs without using all the slots. Viking’s approach allows an SSD to the mix without using any extra space. This is handy for rack-mount servers and offers interesting possibilities for embedded systems. Motherboards with just a pair of DIMM sockets are possible homes for this type of technology.
Hybrid drives like Seagate’s Momentus XT (see “Seagate Delivers Second-Generation Hybrid Hard Drive") and Samsung’s FlashOn will have enough field presence to get a better handle on their efficiency compared to all solid-state storage.
Continue on next page
FRAM will remain tied to microcontrollers, but MRAM and phase-change memory (PCM) are another matter. This year will be very interesting for these two technologies as large-scale delivery of these memory devices begins.
Shipments of high-density memory like Crocus Technology’s MRAM-based Spin Transfer Torque technology will begin to challenge flash memory’s dominance (see “STT Technology Puts A New Spin On MRAM”).
It will have to share the field with PCM, though, which will be coming on strong this year as well (see “Phase-Change Technology Enters The Memory Market”). Both face significant challenges but with the advantage of better features, performance, and capacity.
High-Speed Standards
Faster storage and HD multimedia content that may even be augmented with 3D support are pushing adoption of the latest high-speed serial interfaces like USB 3.0 and SATA-III. These technologies were included on high-end motherboards like Gigabyte’s GA-P55A-UD5 last year but are standard fare these days (see “The Right Combination Delivers The Best Desktop Performance”).
The availability of high-speed, external hard drives like Seagate’s USB 3.0 BlackArmor PS 110 will force motherboard designers into USB 3.0 support (see “USB 3.0: A Tale Of Two Buses”). USB 3.0’s backward compatibility allows the drive to be used with older hosts.
The PCI Express 3.0 standard is finally out, and hardware will begin trickling out early this year (see “PCI Express 3.0 Nears The Station”). A flood is likely to follow as its advantages take hold.
PCI Express 3.0 is more efficient with 128b/130b’s lower encoding overhead while providing twice the throughput at 8 Gbits/s per lane. It also supports data reuse hints to indicate that data should be placed in cache as well as main memory, dynamic power adjustment mechanisms, loose transaction ordering, and atomic operations.
PCI Express will continue to give InfiniBand, Serial RapidIO, and Ethernet a challenge on the backplane. Still, InfiniBand and Serial RapidIO have well established and rather large niches such as HPC and communications, respectively.
The challenge comes as PCI Express deployments take advantage of features like single root-I/O virtualization (SR-IOV) and other clustering features showing up in processors, peripheral chips, and switches (see “Multicore Security Processor Handles 40 Gbits/s”). Advanced Switching has not been resurrected, but many of its features are showing up in simpler implementations (see “Advanced Switching It Is”).
Security and Anti-Cloning
Encrypted communications aside, security has often been overlooked on embedded systems where connectivity was secondary. These days, security issues are in the forefront and hardware features such as secure storage and anti-cloning support are high-end options.
The challenge for designers this year will be to identify those features that will provide them with benefits as well as what benefits are worth the cost. The cost not only includes the additional hardware but also the software and management (see “Prepare Your Counterattack Against Counterfeit Parts”).
Continue on next page
Keeping applications secure is often as important as keeping application data secure. Cloning is becoming more common and expensive to combat or ignore. No single approach is appropriate for all applications, but the features for protecting and running encrypted or signed code are now standard from most processor vendors. Trying to discern how much security to employ remains an issue for those new to this side of system design and development.
Dev Kits More Than Just Boards
Last year was host to a variety of development kits that weren’t just one-off boards designed to highlight a new chip. Instead, kits are often modular and targeted to specific applications. These ideas aren’t new, but the depth of support and novelty is. The result is faster startup time, better evaluation of the underlying hardware, and more useful demonstrations than just a few blinking LEDs.
Freescale’s Tower line of development platforms (see “MCU Exhibits Tower Power”) includes hardware and a full suite of software development tools in addition to the MQX real-time operating system (Fig. 7).
The 16-bit MSP-430 in the EZ430 Chronos watch (see “Low-Cost Kits Make Evaluation Faster”) from Texas Instruments has RF and accelerometer support that can be used to control the TI Stellaris Evalbot kit that runs a 32-bit Stellaris microcontroller (see “Transforming Circuit Board Into A Robot”). Both provide a platform that is closer to a final application solution, leaving developers to concentrate on adding hardware and software more specific to their end product (Fig. 8).
Did I miss something? Definitely. The turn in the economy is slow, but digital electronics continues to change at an accelerated rate. Not all technologies will survive in the long run, but many will, like PCI Express. It’s going to be a fun year for digital developers.