28-03-2011, 09:35 AM
chameleonchipppt.ppt (Size: 1.18 MB / Downloads: 86)
1.Introduction
A reconfigurable processor is a microprocessor with erasable hardware that can rewire itself dynamically.
This allows the chip to adapt effectively to the programming tasks demanded by the particular software they are interfacing with at any given time.
Reconfigurable processor chip usually contains several parallel processing computational units known as functional blocks.
While reconfiguring the chip, the connections inside the functional blocks and the connections in between the functional blocks are changing,
that means when a particular software is loaded the present hardware design is erased and a new hardware design is generated by making a particular number of connections active while making others idle.
This will define the optimum hardware configuration for that particular software.
It takes just 20 microseconds to reconfigure the entire processing array.
Reconfigurable processors are currently available from Chameleon Systems, Billions of Operations (BOPS), and PACT (Parallel Array Computing Technology).
Among those only Chameleon is providing a design environment, which allows customers to convert their algorithms to hardware configuration by themselves.
2.Multifunction Implementation
In a conventional ASIC or FPGA, multiple algorithms are implemented as separate hardware modules. Four algorithms would divide the chip into four functional areas.
With Reconfigurable Technology, the four algorithms are loaded into the entire reconfigurable Fabric one at a time. First, the entire Fabric is dedicated to algorithm 1; during this processing time, algorithm 2 is loaded into the background place. In a single clock cycle, the entire Fabric is swapped to algorithm 2; during this processing time, algorithm 3 is loaded into the background plane. The entire reconfigurable fabric is dedicated to just one algorithm at a time.
So finally the result is: much higher performance, lower cost and lower power consumption
3.The General Architecture Of Reconfigurable Processor
Machine design supposes that some pins are considered as the configuration inputs and another as data or control inputs and outputs.
A new chip must inside determine the set of the function blocks (FB), which are used to construct the circuit, rules of their interconnections and ways of the input/output connections.
The most important parts are the logic circuits, which configure function blocks according to data in the configuration memory.
The various possible connections between functional blocks are encoded to bits known as Configuration bits. Resulting configuration stream is downloaded into configuration memory through configuration inputs.
Thus, a new Reconfigurable machine is established.
4.Architecture
Components:
32-bit Risc ARC processor @125MHz
64 bit memory controller
32 bit PCI controller
reconfigurable processing fabric (RPF)
high speed system bus
programmable I/O (160 pins)
DMA Subsystem
Configuration Subsystem
5.Reconfigurable Processing Fabric(RPF)
The Fabric provides unmatched algorithmic computation power to Chameleon Chip. It consists of 84,32-bit Data path Units and 24, 16×24-bit Multipliers,Operating at 125Mhz, they provide up to 3,000 16-bit Million Multiply-Accumulates Per Second and 24,000 16-bit Million Operations Per Second.
The fabric is divided into Slices, the basic unit of reconfiguration.
The CS2112 has 4 Slices with 3 Tiles in each. Each tile can be reconfigured at runtime
Tiles contain :
Datapath Units
Local Store Memories
16x24 multipliers
Control Logic Unit
The high-performance 32bit Data path Unit (DPU): The Tile includes seven Data path Units. The DPU is a data processing module that directly supports all C and Verilog operations.
6.Programmable I/O
RCP includes banks of Programmable I/O (PIO) pins which provide tremendous bandwidth.
Each PIO bank of 40 PIO pins delivers 0.5 GBytes/sec I/O bandwidth.
7.Technologies Used In Chip
1. eCONFIGURABLE™ TECHNOLOGY:
This technology reconfigures fabric in one clock cycle and increases voice/data/video channels per chip. As mentioned earlier, each Slice can be configured independently. Loading the Background Plane from external memory requires just 3 µsec per Slice; this operation does not interfere with active processing on the Fabric. Swapping the Background Plane into the Active Plane requires just one clock cycle. with eConfigurable Technology; the four algorithms are loaded into the entire reconfigurable processing Fabric one at a time.