原创 Design of high performance QDR2/DDR2 RAM Controller on Stratix-IV FPGA

2010-10-16 23:20 3117 14 16 分类: 消费电子

Abstract

This article introduced a data capturing technology which used on high performance QDR2 SSRAMDDR2 SDRAM interface. This technology base on several primitives, such as DQS Delay ChainInput Phase Alignment and Half-Rate Input. The PLLs’ dynamic phase shift function also been taken in to account.

 

Introduction

We focus on the physic layer especially the read data path of the interface because this is the difficult point of FPGA implement. Both of the QDR2 and DDR2 are source-synchronized interface. The clock and data is edge aligned at the read path. In this design, The first stage using DQS Delay Chain delayed source-synchronized clock ( dqs or cq) to scan the valid window of DQ. Then, derive the most appropriate delay value and apply it on DQS Delay Chain. The second stage using dynamic phase shifting re-synchronize clock to transfer the read data from source-synchronized clock domain to system clock domain. All of this use the I/O resource of Stratix-III/Stratix-IV, avoid complicated constraints.

 

The write data path

The write data path is somewhat simple, employ the DDIO_OUT primitive. The FPGA Clock in the figure is the 270 degree output of PLL. Feed the Write Data Rise with the MSB of parallel write data, and the Write Data Fall with LSB.


 


The main clock K/K# can be generated by the same method. Tie the Write Data Rise to ‘1’, the Write Data Fall to ‘0’ and the clock port using PLL out 0 degree clock.

 

The read data path

The read data path is different with the ALTERA IP Core and the block RAM not been used. Be fit for low latency and full rate circumstance.

 




 

The figure above shows the I/0 structure of StratixIII / StratixIV and the primitives been used.

1)     The DQS(CQ) capture the input data from pin and translate each double edge data to two bits single edge data. On the way from CQ pin to the clock port of DDIO_IN, the CQ delayed by DQS_DELAY_CHAIN but the figure doesn’t show it.

2)     The single edge data sampled by resync_clk in the Alignment_Synchronous Registers, and then enter the Half Data Rate Registers. The Resync_clk is the common source clock of the sys_clk and they have the same frequency.

3)     The Half_Data_Rate_Registers primitive been used but not its intrinsic function. All of the registers working at the same frequency as QDR2 SSRAM.

4)     The read data enter LUT configured selector once they get out of the I/O registers. The sysclk_phase_sel derived from the phase difference of resync_clk and sys_clk.


 

 

文章评论2条评论)

登录后参与讨论

用户1025715 2011-2-18 13:07

ding yi xia!

用户1575195 2010-10-16 23:04

自己顶一个!
相关推荐阅读
用户1575195 2011-10-11 22:15
从10M到100Gbps,以太网芯片接口(二)
XGMII是万兆位以太网中标准的MAC芯片和PHY芯片接口,数据位宽为32bit,工作在156.25MHz双沿采样。它使用HSTL电平标准,在PCB上可以有效传输7CM。因为单板上布局布线的复杂性...
用户1575195 2011-10-11 22:04
从10M到100Gbps,以太网芯片接口(一)
标准的10M/100M以太网MAC芯片与PHY芯片间的接口是MII(介质无关接口)。MII是一个4bit位宽,频率为25MHz的接口。一个标准的MII接口需要16个管脚。它有2个变种,以便在多端口...
用户1575195 2011-09-21 17:40
背板以太网与高能效以太网标准
背板以太网标准 802.3ap标准定义了背板应用的物理层。这些规范包括1Gbps的1000Base-KX,10Gbps使用4条3.125Gbps链路的10Gbase-KX4和10Gbps单条...
用户1575195 2011-09-14 21:58
DDR3的效率杀手——时间参数
DDR系列SDRAM存储芯片的高速率、高集成度和低成本使其理所当然成为存储芯片的一霸。在PC和消费电子领域当然是不必说,它被称为“主存”。其实,随着通信设备价格战愈演愈烈,在看起来水有点深的通信设...
用户1575195 2011-05-13 19:29
互联网又泡沫了吗?
随着国内几家互联网企业扎堆上市,担忧互联网再度泡沫化的声音在互联网上也渐渐多了起来。我比较认同一种观点,在早已多元化的今天,通信行业整体的泡沫化应该还远没到来,可能泡沫化的是某些互联网商业模式。 ...
用户1575195 2011-04-20 21:14
28nm芯片工艺, Xilinx和Altera各执一词
近期, Xilinx和Altera分别在公司召开了28nm工艺新器件发布会。与技术研究该有的严谨不同,对技术成果的展示可以生动活泼。生动活泼之中两家含沙射影的“攻击”让人感叹。正如她们的发布会礼品都选...
我要评论
2
14
关闭 站长推荐上一条 /2 下一条