Embedded Configurable Memories in FPGAs:

Steve Oldridge, Ernie Lin, Jason Clifford, William Ho, Steve Wilton, Jonathan Rose, Zvonko Vranesic


In a current project, we are investigating the architectures and algorithms of FPGAs with on-chip storage. The constant improvement in FPGA and process technology is changing the way FPGAs are used and designed. In the past, the relative low capacities of these devices have limited their use to small logic subcircuits, but as FPGAs grow, designers are able to use them to implement much larger systems. These larger systems look significantly different than small logic subcircuits; one of the key differences is that these systems often contain memory. Memory, therefore, will be a crucial component of future FPGAs.

One of the challenges when embedding memory arrays onto an FPGA is to provide enough interconnect between the memory arrays and the logic resources. The design of a good memory/logic interface is critical. Since memory access time is often the performance bottleneck in many systems, it is crucial that the memory/logic interface provides a flexible high-speed link between logic and memory. If the interface is not flexible enough, many circuits will be unroutable, while if it is too flexible, it will be slower and consume more chip area than is necessary. Much of our research has focused on the design of a good memory/logic interface. We concentrate on two issues: how flexible the interconnect must be, and how the interconnect can be enhanced by providing efficient connections between the memory arrays themselves.

We are also investigating an alternative architecture, in which shallow wide memories are provided by allowing the user to write and read the configuration memory in unused switch blocks. Implementing this technique on a 100 x 100 logic block FPGA with 128 tracks per channel gives a total of 9.46 Megabits of memory which can be accessed using an arbitrary word width. The circuitry to provide access to the configuration bits in this way consists of 3.58 transistors per bit, and results in a routing speed degradation of 11.6%

The 1998 Douglas R. Colton Medal for Research Excellence was received for a portion of this project.

Funding for this project has been provided by Cypress Semiconductor and the B.C. Advanced Systems Institute.


Publications from this research project:


Back to Steve Wilton's Research Page

Back to Steve Wilton's Home Page