Self-organized Wiring and Arrangement of Responsive Modules (SWARM)

Goal

New methodology for layout design of analog integrated circuits to combine procedural and optimizing automation strategies

Motivation

While layout design of digital integrated circuits has been largely automated through optimization algorithms, suchlike approaches do not find evident industrial acceptance in the analog domain. Instead, analog design still relies on the laborious manual design effort of seasoned layout experts. Although the design flow is enhanced by procedural generators, these are typically limited to very basic circuit components. Thus, the design productivity is still smaller than for digital circuits by several orders of magnitude, which hinders the technological progress of multifunctional microelectronic systems dramatically.

State of the Art

Existing automation approaches can be basically divided into the two automation strategies mentioned above. From an academic point of view, these follow two fundamentally different paradigms: optimization algorithms work top-down and are able to consider design requirements explicitly (if they have been formally described), but not implicitly. In contrast, procedural generators work bottom-up and are able to consider design requirements implicitly (if anticipated by the generator developer), but not explicitly.

Our Approach

Since the strengths and weaknesses of the two automation strategies complement each other in this regard, it is assumed that a combination of both paradigms (bottom-up meets top-down) has much more potential to tackle the analog design problem in its entirety than either optimization-based or generator-based approaches alone. But since the two strategies are complementary, they are –on the other hand– incompatible as well. So, the scientific challenge is rooted in the question: how can the two fundamentally different automation paradigms be married?

The SWARM Methodology

Self-organized Wiring and Arrangement of Responsive Modules (SWARM) is an interdisciplinary methodology addressing the design problem with a decentralized multi-agent system. Its basic principle, similar to the roundup of a sheep herd, is to let responsive layout modules (implemented as context-aware procedural generators) interact with each other in a user-defined layout zone. Each module –being an agent– is allowed to autonomously move, rotate and deform itself, while a supervising control organ successively tightens the layout zone to steer the interaction towards increasingly compact layout arrangements.

Benefit

The procedural layout modules can consider their respective design requirements implicitly, while the control organ functions in an optimizing way and may impose superordinate design requirements explicitly. With this bottom-up-meets-top-down concept, conflicts are resolved via the interaction. Considering various principles of self-organization, SWARM is able to evoke the phenomenon of emergence: although each module only has a limited viewpoint and selfishly pursues its personal objectives, remarkable overall solutions can emerge on the global scale.

Examples

Emergence can be observed in nature, amongst others in a flock of birds: although each bird acts autonomously, this leads to collective swarm behavior. Several examples also illustrate this emergent behavior in SWARM.

The following placement problem (where the modules are allowed to move, but not to rotate or deform themselves) demonstrates, that even optimal solutions can arise from the module interaction.

Initial Constellation
Final Constellation

 

Applying SWARM to place-and-route problems (here: an operational transconductance amplifier) shows that the modules manage to satisfy their local design requirements (e.g., the interdigitated positioning of their devices) as well as global design requirements (such as the explicitly imposed layout contour).

Schematic
Layout with Aspect Ratio 1:1
Layout with Aspect Ratio 3:2
Layout with Aspect Ratio 2:1

Outlook

The SWARM methodology is further advanced with the concept of Smart Particles.