In automobiles, as in workstations and enterprise servers, more and more-powerful central processing units (CPU; processors) are required to handle sophisticated real-time applications. In automotive, these applications include navigation, telematics, safety systems and adaptive control, climate control, and high-end graphics for dashboard display and entertainment. However, fast and high-end CPUs generate lots of heat. This is as much a problem in vehicles as in data centers and workstations. The heat has to go somewhere. In data centers, air conditioning helps mitigate the heat problem. In workstations, heat sinks and fans help. In vehicles, already a harsh, often warm environment, heat mitigation requires innovative packaging, including shielding and heat sinks.
Making CPUs faster doesn’t necessarily help. First, faster CPUs generally consume more power, and therefore run hotter. For instance, increasing clock speeds above 3.0 GHz requires dealing with 150 W or more of heat, which poses other chip and system design problems. Second, the associated memory chips have not kept pace by getting faster. As a result, a computer often spends a lot of its time waiting for the memory system to feed data to the processor. So while the CPU is whizzing around burning energy and creating a lot of heat, it’s not really doing much.
Help is on the way. Rather than increase processor frequency, which would increase the speed of number crunching at the expense of sucking up more power (producing more heat) and requiring additional clocking overhead to run, the latest-generation CPUs are both multicore and multithreaded. By throwing more cores at a computational problem, chip designers can keep the frequency of each core relatively low, with concomitant lower power requirements and heat production. (Heat mitigation was the original reason for multicore processor designs.) The multiple cores together consume about the same if not less power than the single- or dual-core CPUs of just a couple of years ago. Less power consumption means less heat emitted, and a smaller cooling “budget.”
In a multicore processor, the whole pro-cessor isn’t doing nothing but producing heat while it waits for more data. Instead, each core can process different parts of data as system memory feeds those data. Likewise, each core can be assigned to respond to different real-time I/O, effectively producing a faster-responding CPU. Moreover, because each core can handle multiple processing threads, each core can process more data per clock cycle than previous generations of CPUs.
Multicore processors also include memory controllers and bridge chips. Such integration obviates the bottlenecks from having discrete memory subsystems. It also eliminates data caching. Caching leads to nondeterministic (i.e., unpredictable) performance, which just won’t do in electronic steering, brake control, adaptive cruise control, and other types of automotive control processes. Better, these integrated hunks of silicon—containing multiple multithreaded cores, memory, bridges, and more—create a whole device that is physically smaller than the sum of the parts. This is an advantage where space is a premium, such as under the dashboard, in a workstation or laptop, or even in a data center.
Helping to make semiconductor devices smaller is the chip industry trend for making the smallest features on the devices smaller and smaller. Up to recently, most of the industry was building chips with 90-nanometer (nm) process technology. In 2005, Intel Corporation (Santa Clara, CA; www.intel.com) began making chips at 65 nanometers. Now Intel and others are moving to feature sizes of 45 nm. This trend in smaller feature sizes enables chip manufacturers to produce chips that require less power to operate. It also lets them cram more in a given space.
Late last year Intel released a quad-core chip that was initially for gamers and others wanting high-definition graphics (think computer-aided design, computational fluid dynamics, and virtual reality). Now Intel offers a family of quad-cores that are showing up in data servers and affordable workstations for the rest of us. For desktop computers, Intel’s Core 2 Quad Qxxxx and Core 2 Extreme QXxxxx chips have clock speeds ranging from 2.4 GHz to 3.33 GHz, from 2 x 4 MB up to 12 MB L2 cache, and range in cost from about $300 to $1,300 on the street. The Intel’s Xeon server quad-core processors are all 65-nm processors, with clock speeds ranging from 1.6 GHz to 3 GHz, and prices ranging from $350 to $1,250. The quad-core server chips require 80 W to 120 W of power.
In enterprise servers
Back in 1986, Sun Microsystems, Inc. (Santa Clara, CA; www.sun.com) produced its first SPARC processor and in the following year shipped its first SPARC workstation. Just over 20 years later, namely this past summer, Sun announced the “world’s fastest commodity microprocessor”: the UltraSPARC T2. This is a multicore, multi-threaded chip with eight cores at 900 MHz to 1.4 GHz per core, and eight threads per core, yielding a 64-way system on a single chip. It also has dual multithreaded 10-GB Ethernet ports (four times the performance of current network interface cards, without the cost of those cards), on-chip network and security functionality, eight lanes of PCI Express I/O, and quad memory controllers that can deliver more than 50 GB/sec of memory access. The processor can support up to 64 logical domains per processor, which can help the network infrastructure as well as improve application consolidation (such as enterprise resource planning plus product lifecycle management plus customer relationship management plus data mining).
This device belies the moniker “full system on a chip.” Say company officials, this “single piece of silicon reduces cost and increases performance, reliability, and energy efficiency—making it the choice for a diversity of workloads, from networking equipment to high-performance comput-ing or storage devices.”
Part of the CPU’s high performance comes from a different approach to processor design than what that first SPARC chip was based upon: Higher compute performance comes from threading rather than the brute-force methods of increasing clock speed, greatly enlarging caches, creating new packaging (such as 90-nm technology), or balancing performance with power consumption. In fact, the UltraSPARC T2 is powered by less than 95 W (nominal)—fewer than 2 W per thread, requiring one-tenth to one-thirtieth the power consumption ofcompetitive CPU offerings.
Prices for the UltraSPARC T2 start “well below” $1,000.
According to the Telematics Research Group, Inc. (Minnetonka, MN; www.telematicsresearch.com), about 25% of today’s automotive bills of materials (BOM) consist of electronic components; by 2015, those components will make up 35%. In the past, automotive telematics, navigation, and entertainment systems were separate systems. The trend today is for these three systems to share electronics. Such integration reduces the number of BOM items and associated costs, increases the reliability of the embedded electronics, reduces power consumption and heat dissipation, and increases the overall speed of the embedded electronic systems.
Case in point: Last May, Freescale Semiconductor Inc. (Austin, TX; www.freescale.com/automotive/) debuted a microprocessor that integrates control electronics for automotive dashboards, telematics, and multimedia into a single, low-power system-on-a-chip. The MPC5121e has all the electronics to support telematics, global positioning systems, and other navigational aids; Bluetooth, Ethernet, and Wi-Fi data communications; and the audio and video requirements for both the automotive dashboard and entertainment systems. According to Freescale officials, the company’s telematics family of micro-processors will soon include functionality for adaptive cruise control, lane departure warnings, pedestrian and collision avoidance, and heads-up display.
The telematics microprocessor is a 90-nm, 516-pin low-power CMOS measuring 27 mm x 27 mm. It includes three processor cores, semiconductor and mass-memory interfaces, and controllers for several types of peripherals (such as 10/100 Ethernet, PCI 2.3, SATA, USB 2.0, and CAN). The three cores are a 300- to 400-MHz e300 Power Architecture core running at up to 760 MIPS over a 64-bit processor bus, a 32-bit RISC-based multimedia accelerator running at 200 MHz, and a 2D/3D graphics core that supports 3D texturing and shading. An on-chip display controller supports 1,280 x 720 pixel LCD. For mass storage, the microprocessor has both serial and parallel ATA interfaces. The chip conforms to the reliability requirements of the AEC-Q100 standard and TS14969 specification for withstanding harsh environments. The microprocessor can run on real-time operating systems from Green Hills Software, QNX Software Systems, Wind River, and run open-source Linux applications. Suffice to say, there are more specs—all in a microprocessor that generates less than two watts of heat (no fan or heat sink is required) and costs about $1,000 each.
While aimed at the automotive market, the MPC5121e is also suitable for industrial control—harsh and hot environments that require monitoring and processing multiple inputs, providing fast I/O, offering quick and intensive graphics, and interacting with various peripherals complying with a variety of data communications standards.