| Brief Overview
of Advantages of Asynchronous Designs.
Since the clock has served the industry so well for so long, one must have
fair enough reasons to consider dispensing with its services and look forward
towards clock-less designs (self-timed or asynchronous designs).
Several arguments have been used to justify abandoning the known and trusted
synchronous design methodology. Though some of the arguments don't seem to be
wholly convincing, when taken together they make a reasonable case for further
research. Every synchronous designer is burdened with the fundamental
assumption that "all signals are binary and time is discrete. Many are
now realizing that their most difficult problems surround these basic
assumptions. Problems which seriously impact their final results are clock
skew, power consumption, critical path delays, metastability, etc. By
eliminating these restrictions, a design can reach higher level of
performance. Asynchronous design offers an interesting alternative by keeping
the restriction that signals are binary but realizing that time is not
discrete. Asynchronous design is a theoretical realization that warrants some
attention. However, as of now we have to be content with the rather diffuse
arguments. Asynchronous designs claim to yield better results in the following
areas of designer's interest: Power efficiency, Modularity, Performance, Clock
Skew, Concurrency and Metastability. |
| Detailed
Review of its Advantages over Synchronous Designs.
As said above, there are several arguments which aren't wholly convincing
though they make a reasonable case for research together. Perhaps, this is the
reason why much of the work is still carried out in academia. Industries
require rather firmer ground to justify redirecting their development
resources. Lets see if the following arguments can fetch something!!!
- Power Efficiency : Reducing power consumption has become
critical with the advent of battery operated consumer electronics. Design
constraints for this class of product require strict power constraints. In
synchronous circuits, clocks beat away, causing power dissipation whether
or not the circuit is doing useful work. For example, even though the
floating-point unit on a processor might not be used in the current
instruction stream, the unit still is operated by clock. Clock gating and
dividing alleviates this problem, but only on a coarse granularity,
moreover that again needs additional hardware. Asynchronous circuits are
inherently data driven and only uses power for useful work. Though
asynchronous circuits seem to be needing more transitions on the
computation path than synchronous circuits, they generally have
transitions only in the areas involved in the current computation. The
Philips digital audio error detector chip based on the asynchronous design
methodology supports the power efficiency claim. It demonstrates a factor
of five power reduction over its synchronous counterparts built on the
same technology.
- Modularity : Clock is a global signal. Handling multiple
clock domains in synchronous designs is not an easy task. Interfacing two
FSMs that were designed with different clock requirements or operating at
different frequencies is very difficult. Asynchronous designs have clean
interfaces and avoid this problem. The modularity problems with clocked
circuits only arise if design components are ported below the RT level. If
they are ported as behavioral descriptions, a synthesis tool should be
able to generate a new RT structure to suit a new clock period. A
perspective which is often applied is to look to a future where a billion
transistors are available on a chip and to argue that the modularity case
will then be compelling, but even here the problem can be removed by
sufficiently powerful abstraction and synthesis tools.
- Performance : In any synchronous design's pipelined datapath, the data marches through the operation chain (recall we have pipeline registers between any two pipeline stage) under the command of the clock.
Clock period is chosen to be larger than the worst-case delay of each pipeline stage. The throughput of
this is equivalent to the number of data samples processed per second
which in turn is equivalent to the clock rate. In other words, throughput of a pipelined system is directly linked to the worst-case delay of the slowest element in the pipeline. On an average, the delay of each pipeline stage is smaller. The same pipeline could support an average throughput rate that is substantially higher than the synchronous one. Throughput here is proportional to the worst-case delay of the slowest stage multiplied by the number of pipeline stages.
The performance of microprocessors and computer peripheral designs are
heavily effected by internal timing. The term "critical-paths"
evolved to describe signal paths in designs that limit the overall circuit
performance. Every good designer knows that critical-paths if reduced will
directly effect the design's performance (processing speeds). Asynchronous
design's exhibit average-case performance as against worst-case
performance. Clearly, the critical timing of a circuit is enhanced when
comparing average case scenarios to worst case scenarios. Synchronous
designs cannot avoid worst-case performance since all possible
computations must be complete before results can be latched. On the other
hand asynchronous designs are free from the dilemma since the circuitry is
capable of sensing the computations completion.
- Clock Skew : Synchronous designs assume that all clock events or timing references happen simultaneously over the complete circuit. This
in fact is not the reality, because of the "clock skew" effects. Clock skew is fast becoming the single most difficult phenomena to protect a circuit against it.
As clocks get faster, chips get better and tracks get thinner, it is
increasingly hard to keep clock skew within tolerable limits. The
engineering effort and silicon resource that goes into the clock
distribution network on today's high-end processors illustrate the scale
of the problem. Asynchronous designs have no clocks and hence do not
suffer from clock-skew. The performance and clock skew arguments are
related. High performance of the today's high end processors does not
appear in general to be incompatible with power-efficiency, and adding
clock gating to improve power-efficiency compromises clock skew thereby
adversely affects performance, but a direct comparison between high
performance clocked and asynchronous circuits is difficult because of the
imbalance in the resources available to the two methodologies.
- Concurrency : Most abstract
representations of concurrent systems map far more naturally onto
asynchronous circuits than they do onto clocked circuits. This however is
depending on the tools support which is not yet widely available.
- Metastability : Any
asynchronous input (any input from outside world) to a clocked circuit
represents a source of unreliability, since there is always some residual
probability that the synchronization circuit will fail. Metastability wreaks havoc in synchronous systems. It is caused by unstable equilibrium state for example when a pair of cross coupled CMOS inverters are stuck at mid-voltages it is impossible to determine how long such a state persists. Unfortunately, due to the complexities in today's systems, it is not possible for the designer to avoid this type of situation.
Asynchronous systems naturally handle mutual exclusivity since they can wait an arbitrarily long time for such a condition to stabilize. Also since these systems are not bounded by global clock, inputs from external world are accepted more gracefully since they are naturally asynchronous.
- Easing of Global Timing Issues : In
systems such as a synchronous microprocessor, the system clock, and thus
system performance, is dictated by the slowest (critical) path. Thus, most
portions of a circuit have to be carefully optimized to achieve highest
clock rate, including rarely used portions of the system. Since many
asynchronous systems operate at the speed of the circuit path currently in
operation, rarely used portions of the circuit can be left unoptimized
without adversely affecting the system performance.
- Better Technology Migration Potential : Integrated
circuits will be implemented in several technologies during their
lifetime. Greater performance for a synchronous system can be obtained by
migrating all system components to a new faster technology, since the
overall performance is again based on the longest path. In asynchronous
systems, migration of only the more critical system components is possible
and can improve system performance. Also since many asynchronous systems
sense computation completion, components with different delays may often
be substituted into a system without altering other elements or
structures.
- Automatic Adaptation to Physical
Properties : Delay through a circuit can change with the
variations in temperature, power-supply voltage, fabrication and other
physical factors. Synchronous circuits must take into consideration of the
worst possible combination of these factors and thus determine the
clocking speed. But asynchronous circuits sense computation completion,
and will run as quickly as the current physical properties allow.
|
Obstacles to Asynchronous Design
-
Testability : Although one
can argue that power consumption, performance, EMI, etc., of asynchronous
signals are very real, so are the threats. The most frequently cited
argument against asynchronous design is the difficulty of production
testing asynchronous devices. Much less is known about the testability of
such circuits, and design margin characterization poses a significant
problem. Production test equipment is heavily committed to using a clock
to regulate the application of stimuli and to time the sampling of
responses from the device under test. Timing driven test equipment is
particularly not suitable for testing event-driven asynchronous circuits.
In addition, asynchronous circuits tend to require logically redundant
gates to remove hazards, and these are un-testable by normal methods.
Economic testability is an absolute requirement for volume
production.
-
Design Tools Support : Design
tool support is another area of difficulty. Modern CAD tools are aimed at
clocked circuits, and though it is clearly possible to design a clockless
circuit using them, they could be lot more helpful, if the tools were
targetted for clockless designs. For example, most synthesis tools will
automatically remove logically redundant gates that were carefully
inserted to remove hazards. Recent developments however, prove the
feasibility of CAD tools targetted for asynchronous designs, and it is
just a matter of time before large asynchronous designs enter the
commercial arena.
-
Design Practice : Perhaps,
the biggest obstacle to the acceptance of asynchronous design techniques
assuming their merits can be demonstrated irrefutably, is the aversion of
the design community. Engineers have been trained to avoid asynchronous
operation, though it was used in the distant past. This causes that the
designers will be reluctant to suggestions to revert back to asynchronous
design methodologies that were rejected a few of decades ago.
|
