24.
10.
2014
Low Power System Design
Heinrich Theodor Vierhaus
BTU Cottbus
Sondertermin: 11:30 Uhr, INF/3105
Electronic systems of today benefit from ever-increasing density and
functionality of basic circuits. But large-scale integration is now
approaching physical limits, whereby power dissipation is among the most
crucial points. On the other hand, new systems and applications require
amounts of computing power that cannot be realized with existing
technologies due to power requirements. The main question then is where
and how to save power. The tutorial tries to introduce means of saving
power from circuit design via processor architecture to software and
algorithms in order to explore, where power can be saved at what cost. We
start with basic mechanisms of power dissipation in CMOS, look at
innovative logic designs like sub-threshold and adiabatic logic, and end
up with management of clock trees. For processors, we take a look at
design principles and micro-architectures with respect to power
dissipation, finally ending up in more or less power hungry algorithms.
Next interdependencies between fault tolerance, reliability and power are
discussed. Finally, we take a short look at principles of energy
harvesting.
24.
10.
2014
Self Repair Technology for Integrated Circuits
Heinrich Theodor Vierhaus
BTU Cottbus
The down-scaling of feature size in integrated circuit technologies has
yielded an almost exponential growth in performance over the last 45
years. With dimensions now reaching dimensions of 20 nanometers and
below, however, integrated circuits are becoming less reliable than
before, due to a variety of different fault effects. Smaller feature size
and reduced voltage levels make circuits more susceptible to transient
fault effects, for example by particle radiation. On the other hand,
thermal stress promotes aging effects, which results in reliable circuit
operation for much less than the decades shown by older generations of
ICs. Even worse, extra circuits used for transient fault detection and
correction always costs extra power, which reduces circuit life time. In
order to yield long-living dependable systems from not highly reliable
components, built-in self repair (BISR) has been discussed for several
years. BISR uses "fresh" redundant resources for repair and has found
widespread application in embedded memory blocks of large-scale
integrated systems. In the tutorial, we introduce fault mechanisms and
error correction first. Then we explain built-in self repair for logic
structures, based on re-organization by either hardware or software.
|