Dual ports are considered specialty memories-using a larger, eight-transistor cell structure instead of the standard, six-transistor cell common to most SRAMs-so they cost more than standard memories. The advantage of the dual port is that it can be written to and read from through two independent ports running at different speeds and out of phase with each other, allowing, for example, two processors to share the same data set stored in one device.

Alternatively, system designers could use a standard SRAM and allow a processor to access it through a bus topology, but bus contention issues could reduce performance by a factor of three, said Geoff Charubin, director of marketing for Cypress' data communications division.

Cypress thinks the time is right to freshen up its dual-port product line as orders increase. Last quarter the company's sales topped $200 million, up from $181 million the previous quarter.

"We're starting to see a slight pickup in these markets, and there are new systems rolling out that will need these specialty memories," Charubin said.

The total available market for dual-port memories should come in at around $130 million this year, just a fraction of the overall memory market. But that would be up from $110 million last year, and Cypress is in position to share that market with just one other major vendor-market leader Integrated Device Technology Inc., behind which Cypress says it places a close second.

What makes Cypress' latest dual port stand out are its large density and wide I/O interface. The FLEx72 chip can hold 18 Mbits of data, twice as much as Cypress' next-highest-density dual port. There's a pressing need for more density in systems like basestations, and that calls for fast, shared memories that sit between a pair of DSPs or between an FPGA and a DSP, said marketing manager Brent Roberts.

Cypress thinks storage applications can also benefit from larger dual-ported memories. Storage nets make extensive use of "credits," which are similar to frames and must be acknowledged and received before data can pass from node to node. As the networks expand, so does the transmission time for the credits. That can be mitigated via memory devices that can store more credits concurrently, Roberts said.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved