| |
The
main goal of the project is to develop the HW/SW intellectual
property (IP) cores and platforms for implementing a general-purpose
associative dynamic memory for storing and retrieving binary
feature descriptors in Silicon platforms. This includes
creating the general IP and system framework. Beyond developing
an abstract hardware/software hybrid model supporting sophisticated
pattern recognition and search tasks, the projects also
targets a system design approach for a few concrete application
fields, as required by SME business domains. The project
will deliver a Demonstrator for content-based text retrieval
system. It includes the corresponding IP core design and
exploitation methodology, the system software, and the
middleware resources needed by the full application for
the three different companies and application domains.
Based on this general architecture, AMASS will produce
application specific platforms for database, multimedia
and internet semantic search.
The hardware platform does not imply a fixed silicon implementation.
In this project, both hardware and software processes will
be mapped into prototyping platforms in order to determine
the optimal balance between performance and cost, keeping
the option to change or update both reconfigurable hardware
and programmable software. This is possible because FPGA
chips allow for downloading hardware, software, or even
processor code. The price of these devices, counted per
logic gate and Megahertz, is very competitive for devices
requiring less that 100.000 units per year, as in our case.
Furthermore, FPGA development platforms provide the fastest
and most cost-effective entry on ASIC or standard technology
mass-products by directly mapping functional entities into
silicon structures.
Short Technical Description
The AMASS project seeks to develop a rather universal platform
on which the three SME participants can build their own
applications in only slightly overlapping application domains:
knowledge base search, personalized internet profiling,
and multimedia management. The AMASS project is a short-medium
term project: after two years, we wish to sell the product.
Experiments
on a FPGA platform show that the speedup factor
compared to general purpose mainframes
is more
than 50 when accessing standard SRAM internal memory,
due to FPGA ability to address internally storage devices
via parallel, wideband channels. This factor is on
top of a technology growing itself under Moore’s
law. Therefore, there exists a realistic chance
of developing
a sustainable, competitive SW/HW system-on-chip technology
that will drive both the evolution of existing SME
applications as well as open new opportunities
and markets. Given
the predominance of East-Asian and US manufacturers
on the storage media market, the AMASS platform(s)
will
be also desirable as an intellectual property asset
per se. In our interpretation, this is one of the
main goals
of the CRAFT programme. |
Since the main goal of AMASS is to move the processing of
the data index away from the processor and close to the
storage media, the platform can be used as an entry gate
to any medium, including magnetic disks, RAM, or flash
storage modules. Using magnetic disks for internal memory
seems a very competitive alternative to SDRAM, because
the storage medium is cheap (250 GB for 150 Eu) and compact
(mini-disks).
You
can read more about
Search
Engines for pedestrians
The
AMASS platform
in
this PDF document 
|
|
