Copyright © 2006 Penton Media, Inc., All rights reserved.
Printing of this document is for personal use only.
Reprints

Download the full article as a .PDF, sponsored by Linear Technology.

Why do FPGAs consume so much power?
Historically, FPGAs have had a more passive “glue logic” role hardly worth mentioning in the overall power budget. But now they have taken over most of the system control, and today’s FPGAs can be a system-on-a-chip (SoC) solution. That means they also consume a greater percentage of the total system power budget. But you can do several things at the chip and system levels to help meet your power budget.

What are the latest silicon innovations that help reduce power consumption?
As we move into deep-submicron geometries, increases in functionality per square millimeter come at the cost of higher static power consumption due to higher transistor leakage. Supplying the FPGA core voltage at the lower limit of the manufacturer’s specification can save significant static power. For example, static power might increase 15% for a mere 5% increase in core voltage (see the figure).

Some FPGA vendors use a triple-oxide process technology for some transistors to reduce static power consumption of non-speed-critical configuration circuitry. Another innovation is a shift to a coarser-grained logic architectures employing lookup tables (LUTs) with six inputs rather than the previous standard of four. This enables tighter logic packing, reduces the number of switching transistors, and shortens routing lengths, reducing dynamic power consumption.

How can power consumption be reduced at the chip level?
One of the more powerful techniques is to maximize the use of hard IP blocks available on chip, because FPGA vendors design the hard IP to use only the exact resources required to achieve a given protocol or architecture.

Another technique is to simply suspend all or part of the FPGA when it’s not in use. Putting the FPGA “to sleep” when it’s not in use is well understood. But even when it is in use, its ability to “shut off” parts that aren’t needed for a given function often is overlooked. Internal RAM should be disabled when it’s not in use. Use gated and/or local clocking to minimize power dissipation in the clock networks.

Also, take care when writing HDL code. Implement techniques like one-hot encoding for state machines and guarded evaluation while making sure you don’t inadvertently create logic loops that oscillate. And finally, choose your I/O standards wisely, as some consume more static power than others.

How can tools help reduce FPGA power consumption?
One common mistake is to overconstrain FPGA timing or provide no timing constraints (whereby the tools may assume to try and build the fastest implementation possible). This leads the synthesis and place & route (P&R) tools to make every attempt to meet the constraints, normally leading to unnecessary widening of logic blocks and duplication of logic and registers. It also discourages the tools from sharing logic resources. The same goal often can be achieved by specifying the maximum allowable timing slack, encouraging tools to optimize for smaller area and power.

What can be done to reduce power consumption at the board level?
For optimal power consumption, consider these factors:

• The device operating (junction) temperature: Reduce the junction temperature to reduce static power and increase device reliability and lifespan.
• Airflow: To reduce junction temperature, implement correctby- design airflow. Allow for proper airflow around the FPGA by choosing low-profile regulators that won’t cause thermal shadowing. This will allow for less expensive cooling solutions, such as smaller heatsinks, if any.
• Package spacing: Packages need to have proper spacing, with special consideration for keeping maximum distance between heat-producing parts.
• Package type: The choice of package can have a dramatic impact on power efficiency. If you can use a package like an LGA that solders directly to the PCB, the PCB then becomes the primary source of heat dissipation for that part.
• Efficient voltage conversion: To reduce power wasted in the power regulators, use converters with efficiency ratings of 90% and higher.

Linear Technology offers DC/DC µModule™ regulators ideally suited for Virtex-5 FPGA-based boards such as this ML525 RocketIO™ serial transceiver characterization platform with XC5V330T, the world’s largest FPGA.
The LGA package used on the Linear Technology µModule regulators allows for excellent heatdissipation characteristics.	The ultra-compact size of the Linear Technology µModule regulators enables designers to maximize board-space savings.

Product Q&As

VirtexTM-5 FPGAs Enable Complex System Integration

Xilinx® Virtex™-5 FPGAs are fabricated on a 65-nm, triple-oxide process to deliver the highest performance with the lowest power. Built-in IP blocks including PCI Express® endpoint, Gigabit Ethernet MAC, DSP engines, and serial transceivers increase integration and reduce power consumption.

Complete Solutions Accelerate System Design
The Xilinx design ecosystem, including development kits, design tools, pre-verified IP, and a worldwide infrastructure of training and services, provides everything you need to start designing with Virtex-5 FPGAs. Start designing today: visit www.xilinx.com/virtex5 to download design tools, including the free XPE power estimator.

Linear Technology’s family of DC/DC µModule regulators simplifies Xilinx-based system board design. These complete DC/DC systems come with on-board inductor, MOSFETs, DC/DC controller, input and output bypass capacitors, as well as compensation circuitry. Although the µModule regulator can deliver high output power, its excellent thermal characteristics allow it to be placed near the rest of the ICs. Also, its small size and low profile enable digital system designers to make the most of valuable board space without compromising performance. With seven members of the µModule regulator family already released and more scheduled for 2008, customers can find solutions from 4 A up to 12 A. The µModule regulators can also be easily paralleled together up to 48 A. Find out more at www.linear.com/xilinx.