OpenRAM/ICCAD16_openram_paper/introduction.tex

69 lines
3.6 KiB
TeX

\section{Introduction}
\label{sec:introduction}
% why memory compilers are important
Static Random Access Memories (SRAMs) have become a standard component
embedded in all System-on-Chip (SoC), Application-Specific Integrated
Circuit (ASIC), and micro-processor designs. Their wide application
leads to a variety of requirements in circuit design and memory
configuration. However, manual design is
too time consuming. The
regular structure of memories leads well to automation that produces
size and configuration variations quickly, but developing this with
multiple technologies and tool methodologies is challenging. In
addition, memory designs play a significant role in overall system
performance and costs, so optimization is important. Thus, a memory
compiler is a critical tool.
% why academics need memory compilers
Most academic ICs design methodologies are limited by the availability
of memories. Many standard-cell Process Design Kits (PDKs) are
available from foundries and vendors, but these PDKs frequently do not
come with memory arrays or memory compilers. If a memory compiler is
freely available, it often only supports a generic process technology
that is not fabricable. Due to academic funding restrictions,
commercial industry solutions are often not feasible for
researchers. In addition, these commercial solutions are limited in
customization of the memory sizes and specific components of the
memory. PDKs may have the options to request \enquote{black box}
memory models, but these are also not modifiable and have limited
available configurations. These restrictions and licensing issues make
comparison and experimentation with real world memories impossible.
% manually designing is time consuming
Academic researchers are able to design their own custom memories, but
this can be a tedious and time-consuming task and may not be the intended
purpose of the research. Frequently, the memory design is the bare
minimum that the research project requires,
and, because of this, the memory designs are often inferior and are not
optimized. In memory research, peripheral circuits are often not
considered when comparing memory performance and density. The
lack of a customizable compiler makes it difficult for researchers to
prototype and verify circuits and methodologies beyond a single row or
column of memory cells.
% what are the goals of OpenRAM
The OpenRAM project aims to provide an open-source memory compiler
development framework for memories. It provides reference circuit and
physical implementations in a generic $45$nm technology and fabricable
Scalable CMOS (SCMOS), but it has also been ported to several
commercial technology nodes using a simple technology file. OpenRAM
also includes a characterization methodology so that it can generate
the timing and power characterization results in addition to circuits and
layout while remaining independent of specific commercial tools. Most
importantly, OpenRAM is completely user-modifiable since all source
code is open source at:
\begin{center}
\url{https://openram.soe.ucsc.edu/}
\end{center}
The remainder of this paper is organized as follows:
Section~\ref{sec:background} provides a background on previous memory
compilers. Section~\ref{sec:architecture} presents the reference
memory architecture in OpenRAM. Section~\ref{sec:implementation}
specifically introduces the implementation and main features of the
OpenRAM memory compiler. In Section~\ref{sec:results}, an analysis of
the area, timing and power is shown for different sizes and
technologies of memory. Finally, the paper is summarized in
Section~\ref{sec:conclusions}.