Visible to Intel only — GUID: iga1463598419039
Ixiasoft
Visible to Intel only — GUID: iga1463598419039
Ixiasoft
48.3. Functional Description
The Avalon® MM DDR Memory Half Rate Bridge works under two constraints:
- Its memory-side host has a clock frequency that is synchronous (zero phase shift) to, and twice the frequency of, the CPU-side agent.
- Its memory-side host is half as wide as its CPU-side agent.
The bridge leverages these two constraints to provide lightweight, low-latency clock-crossing logic between the CPU and the memory. These constraints are in contrast with the Avalon® -MM Clock-Crossing Bridge, which makes no assumptions about the frequency/phase relationship between the host- and agent-side clocks, and provides higher-latency logic that fully-synchronizes all signals that pass between the two domains.
The Avalon® MM DDR Memory Half-Rate Bridge has an Avalon® -MM agent interface that accepts single-word (non-bursting) transactions. When the agent interface receives a transaction from a connected CPU, it issues a two-word burst transaction on its host interface (which is half as wide and twice as fast). If the transaction is a read request, the bridge's host interface waits for the agent’s two-word response, concatenates the two words, and presents them as a single readdata word on its agent interface to the CPU. Every time the data width is halved, the clock rate is doubled. As a result, the data throughput is matched between the CPU and the off-chip memory device.
The figure below shows the latency in the Avalon® -MM Half-Rate Bridge core. The core adds two cycles of latency in the agent clock domain for read transactions. The first cycle is introduced during the command phase of the transaction and the second cycle, during the response phase of the transaction. The total latency is 2+<x>, where <x> refers to the latency of the DDR SDRAM high-performance memory controller. Using the clock crossing bridge for this same purpose would impose approximately 12 cycles of additional latency.