Spartan3e SPI Master配置在大约5-10％的时间内失败

我有两个定制板，每个都有一个采用TQ144封装的Spartan3e器件（一个是100kilogate，另一个是250kilogate，否则电路板是相同的）。
它们已设置为使用“FAST READ”从Atmel AT45DB081D闪存芯片配置SPI，（HSWAP =，M [2：0] =，VS [2：0] =）。
比特流存储为.mcs文件（配置速率设置为“1”），并通过Xilinx平台电缆USB II上的Impact直接加载到闪存芯片中。
配置在大多数情况下都能正常工作，但在大约5-10％的情况下，配置失败（DONE不会变高）。
两块板都安装在同一个多槽背板上，并使用相同的电源供电 - 两块板都是相同的，也就是说，它们都配置成功，或者都失败了。
两块板都采用5V电源供电，并通过板载电源创建自己的3.3V，2.5V和1.2V电压。
2.5V和1.2V电压由3.3V电源产生。
所以：
- > 2.5V
/
5V  - > 3.3V  -  

- > 1.2V
我自己的怀疑是：
1）Flash SPI时钟信号的错误路由：电路板的初衷是允许FPGA和板载MCU访问Flash芯片，因此SPI信号已经路由到FPGA和MCU。
我可以在Spartan3e系列数据表（DS312）中看到，应该注意将时钟信号路由为传输线 - 这还没有完成。
但是，如果这是问题，我会发现失败比它们更随机，而且并不总是同时发生在两张卡上？
2）VS引脚上的错误上拉：配置完成后，VS引脚用于与其他板载设备通信（HSWAP和M [2：0]是专用的，配置后不使用），我可以看到
在配置期间创建'1'的上拉电路实际上由不同的3.3V电源供电，而不是连接到FPGA的电源（虽然接地仍然连接）。
但是，我尝试将HSWAP拉低，以便在其中一块电路板上启用内部上拉，但它对故障没有任何影响？
两张卡（一张HSWAP = 0，一张HSWAP = 1）仍然同时失败。
3）闪存芯片在被选中之前太迟供电电压：DS312说这可能是一个问题，因为只要VCCO_2（3.3V），VCCINT（1.2V）和VCCAUX（2.5V）变高，FPGA就会开始配置，
并且由于Flash芯片需要在被选中前一段时间供电。
但是，从数据表中，我可以看到Atmel AT45DB081D闪存具有70 us的TVCSL（VCC（最小值）到芯片选择低）。
它由3.3V电源供电，2.5V和1.2V电源均由3.3V电源产生，因为这些电源（根据数据表）最多需要250 us才能上升（这就是
比目前板上的输出电容小得多，应该有足够的时间让Flash启动？
我今天晚些时候会对此进行一些测量。
是否有其他人对可能出现的问题有任何意见或其他想法或经验？
有什么我想念的吗？
提前致谢！



以下为原文

I have two custom made boards, each with a Spartan3e device in a TQ144 package (one is a 100kilogate, the other is a 250kilogate, otherwise the boards are the same). They have been setup to configure over SPI from an Atmel AT45DB081D flash chip using "FAST READ", (HSWAP = <1>, M[2:0] = <0:0:1>, VS[2:0] = <1:1:1>). The bitstream is stored as an .mcs file (with configuration rate set to '1'), and loaded directly into the Flash chip through Impact over a Xilinx Platform Cable USB II.

Configuration works fine most of the time, but in about 5-10% of the cases, configuration fails (DONE does not go high). Both boards are mounted in the same multi-slot backplane, and powered from the same power supply - and both boards do the same, that is, either they both configure successfully, or they both fail.

Both boards take in a 5V supply voltage, and create their own 3.3V, 2.5V and 1.2V voltages through onboard power supplies. The 2.5V and 1.2V voltages are created from the 3.3V supply. So:

               -> 2.5V             / 5V -> 3.3V -                            -> 1.2V


My own suspicions have been:

1) Bad routing for the Flash SPI clock signal: The original intention of the board was to allow both the FPGA and an onboard MCU to access the Flash chip, and thus the SPI signals have been routed to both FPGA and MCU. I can see in the Spartan3e family datasheet (DS312) that care should be taken to route the clock signal as a transmission line - which hasn't been done. However, if this was the problem, I'd expect the failures to be more random than they are, and not always happen to both cards at the same time?

2) Bad pull-ups on the VS pins: After configuration, the VS pins are used to communicate with other onboard devices (HSWAP and M[2:0] are dedicated, and not used after configuration), and I can see that the pull-ups creating the '1's during configuration are actually powered by a different 3.3V power supply, than the one connected to the FPGA (grounds are still connected though). However, I've tried pulling HSWAP low to enable internal pullups on one of the boards, but it didn't do any difference to the failures at all? Both cards (one with HSWAP = 0 and one with HSWAP = 1) still fail at the same time.

3)  Flash chip getting its supply voltage too late before being selected: DS312 says this might be a problem, as the FPGA starts configuration as soon as VCCO_2 (3.3V), VCCINT (1.2V) and VCCAUX (2.5V) go high, and as the Flash chip needs to be powered some time before being selected. However, from the datasheets, I can see that the Atmel AT45DB081D Flash has a TVCSL (VCC(min.) to Chip Select low) of 70 us. It is powered by the 3.3V supply, and as both 2.5V and 1.2V supplies are generated from the 3.3V supply, and as these supplies (according to the datasheet) should take up to 250 us to ramp up (and that's with a much smaller output capacitor than is currently on the boards), there should be time enough for the Flash to start up? I'm going to do some measurements on this later today though.


Do anyone else have comments or other ideas or experiences as to what could be wrong? Is there anything I'm missing? Thanks in advance!