Detail publikace

Multifunctionality matters: Towards better implementation of reconfigurable computational elements

NEVORAL, J. RŮŽIČKA, R.

Originální název

Multifunctionality matters: Towards better implementation of reconfigurable computational elements

Typ

různé

Jazyk

angličtina

Originální abstrakt

Nearly 20 years ago, a non-conventional approach to implementation of multifunctional circuits called Polymorphic electronics was proposed. The concept of polymorphic electronics allows to implement two or more functions in a single circuit, whereas the currently selected function depends on the state of the circuit operating environment. Key components of such circuits are polymorphic gates. Since the introduction of polymorphic electronics, few tens of polymorphic gates, mainly controlled by chip temperature or by level of power supply voltage, have been published. However, large number of those gates exhibit parameters that fall behind ubiquitous CMOS technology. As a result of that, perspective of their utilisation for real applications becomes rather bleak. This paper proposes a new approach to the polymorphic electronics. It is based on gates whose behaviour depends on polarity of dedicated power supply rails. Such approach allows to design gates with significantly improved parameters. In order to systematically design those gates at the transistor level, an evolutionary approach based on Cartesian genetic programming is proposed. Using the evolutionary approach, several sets of bi-functional polymorphic gates were designed and validated by HSPICE simulations. They are now available to the research community in a freely available library. Space-efficiency of newly designed gates is demonstrated at two higher levels of abstraction -- bi-functional RTL components and applications (bi-functional image filters). For this purpose, procedure involving conventional synthesis tool and evolutionary optimisation based on Cartesian genetic programming was utilised. Both designed RTL components and image filters show to be significantly area efficient compared to the current solutions.

Autoři

NEVORAL, J.; RŮŽIČKA, R.

Vydáno

31. 12. 2019

Strany počet

28

BibTex

@misc{BUT162284,
  author="Jan {Nevoral} and Richard {Růžička}",
  title="Multifunctionality matters: Towards better implementation of reconfigurable computational elements",
  year="2019",
  pages="28",
  note="miscellaneous"
}