High-speed processor core comprising direct processor-to-memory connectivity
View Patent ↗A high-speed processor core having a plurality of individual FPGA-based processing elements configured in a synchronous or asynchronous pipeline architecture with direct processor-to-memory interconnectivity that avoids the latency and bus contention delays of FPGAs using conventional bused memory. The high-speed processor core has one or more memory structures such as SDRAM or QDR electronic memory and is electrically coupled directly to one or more FPGAs using an access lead network to provide a pipelined set of FPGA-based processor elements for processing one or more predetermined operations such as one or more detection algorithms at line rates.
1. An electronic processor core comprising:
a first reconfigurable processing element configured to execute a first predetermined algorithm on input data and to output processed data from the first predetermined algorithm having a variably wide word width as an output data set,
a second reconfigurable processing element configured to execute a second predetermined algorithm,
the first processing element and the second processing element configured so that the variably wide word width output data set of the first processing element is received as a variably wide word width input data set of the second processing element,
the first and second processing elements each comprising a processor, an access lead network electrically coupled and proximate to the processor and a plurality of external memories electrically coupled and proximate to the access lead network, and,
wherein the processor can independently access each of the plurality of external memories via the access lead network without use of an address/data bus.
2. An electronic processor core comprising:
a first reconfigurable processing element configured to execute a first predetermined algorithm on input data and to output processed data from the first predetermined algorithm having a variably wide word width as an output data set,
a second reconfigurable processing element configured to execute a second predetermined algorithm,
the first processing element and the second processing element configured so that the variably wide word width output data set of the first processing element is received as a variably wide word width input data set of the second processing element,
the first and second processing elements each comprising a field programmable gate array, an access lead network electrically coupled and proximate to the field programmable gate array and a plurality of external memories electrically coupled and proximate to the access lead network, and,
wherein the field programmable gate array can independently access each of the plurality of external memories via the access lead network without use of an address/data bus.
3. The device of claim 2 wherein the field programmable gate arrays are arranged and configured to operate with a variable word width.
4. The device of claim 2 where the field programmable gate arrays are arranged and configured to operate with a word width between 1 to mXN bits where m is the number of bits in the word width of each memory and N is the number of memories.
5. The device of claim 2 wherein the first processing element and the second processing element are configured in an asynchronous pipeline architecture.
6. The device of claim 2 where at least one of the memories is a DDR SDRAM memory.
7. The device of claim 2 wherein at least one of the memories is a QDR SDRAM memory.
8. The device of claim 2 wherein the inputs and outputs of a plurality of the processing elements are configured in a matrix arrangement.
9. A method for processing a data set comprising:
providing a first reconfigurable processing element configured to execute a first predetermined algorithm and a second reconfigurable processing element configured to execute a second predetermined operation algorithm,
the first processing element and the second processing element configured in an asynchronous pipeline architecture whereby an output data set of an algorithm executed by the first processing element is received as the input data set of the second processing element,
the first and second processing elements comprising a field programmable gate array, an access lead network electrically coupled and proximate to the field programmable gate array and a plurality of external memories electrically coupled and proximate to the access lead network,
wherein the field programmable gate array can independently access each of the plurality of external memories via the access lead network without use of an address/data bus,
performing a first predetermined operation on a primary data set using the first processing element to generate an output data set,
inputting the output data to the second processing element, and
performing a second predetermined operation on the output data set using the second processing element.
10. The method of claim 9 wherein the field programmable gate arrays are arranged and configured to operate with a variable word width.
11. The method of claim 9 where the field programmable gate arrays are arranged and configured to operate with a word width between 1 to mXN bits where m is the number of bits in the word width of each memory and N is the number of memories.
12. The method of claim 9 wherein the first processing element and the second processing element are configured in an asynchronous pipeline architecture.
13. The device of claim 9 where at least one of the memories is a DDR SDRAM memory.
14. The device of claim 9 wherein at least one of the memories is a QDR SDRAM memory.