|
|
|
|
|
|
|
| Jeff Bier’s Impulse Response – Faster signal processing: a double-edged sword? |
By Jeff Bier, 4/1/2007
Recent years have seen a proliferation of architectures for digital signal processing. Running the gamut from general-purpose processors and application-specific SoCs to the more-recent DSP-enhanced FPGAs and multi-core processors, these architectures promise unprecedented DSP performance. This has left me wondering: Is more speed always a good thing?
Increased DSP throughput over previous architectures does mean that more tasks are processed in a given time, with a given set of resources. Hardware executing a task—multipliers, logic units, memory—can be switched to a different task sooner, allowing for more efficient utilization of those resources. And this has positive implications for energy efficiency and cost, as well.
This increased throughput has a downside, however. Faster switching of hardware resources intensifies electron momentum effects in those resources. When the electron momentum exceeds a critical threshold, electromigration effects become significant, causing wear and reducing chip lifetime. Research suggests that current architectures are very close to this threshold.
If so, the picture isn’t rosy. Chip vendors, in their pursuit of unbridled DSP performance, seem to have ignored a crucial factor: durability. While high throughput is critical for many applications, durability can be even more important, especially for mission-critical applications. Vendors can simply not afford to ignore it.
How might processor developers avoid a crisis of reliability while continuing to offer increased throughput? Lubricating hardware components (as is typically done with mechanical parts) seems to be a promising approach. Recently, researchers at the University of East Texas discovered a new class of subatomic particle, the anti-gluon, which lubricates the movements of electrons and protons. By doping integrated circuit conductors with anti-gluons, the researchers believe that they can reduce friction within these conductors and effectively check electromigration.
Today, the faster a chip is, the sooner it’s likely to fail. Clearly, this cannot continue. Vendors fielding powerful architectures must choose between addressing the impending reliability crisis head-on and facing extinction.
|
|
|
|
|
| |
|
|
|
|
|
|
|