Cyclone V PLL : "Normal" mode compensation is not working.

It seems I have encountered a Problem with PLLs in Cyclone V devices,

when the PLL is operating in "normal" compensation mode.

According to the Cyclone V manual, a PLL in normal mode should compensate the

delay of a global clock network in such a way, that the clock routed via the global

network should reach all the destination registers (almost) co-incident to the

external reference input of the PLL.

For Cyclone 1 and Cyclone IV devices, that seems to work as expected.

But for Cyclone V devices, the TimeQuest analysis says that this does not work at all.

Even worse some tests with real Cyclone V hardware suggest that TimeQuest is not

(completely) wrong: It seems that somehow the "normal" compensation mode does

not work as expected.

I put two mini example designs for Cyclone IV and Cyclone V here:

http://www.lauterbach.com/support/static/altera_tco_tests.zip

Both designs were compiled with Quartus II 16.1.0 (Build 196 10/24/2016).

My guess is, that the Quartus II version, does not really matter

(I expect the problem to appear with any Quartus II version).

For both designs you can get a timing analysis, by opening TimeQuest

and then running "Script->report_timing.tcl".

I am particularly interested in the

slow and fast corner "WITH PLL" panels.

The part of the design I am interested in does this:

                GLOBAL
                CLK NET
  iClk -> PLL ----------> I/O Register ----> Pin
                          rDataB             oDataB

I am measuring the delay from a rising edge on

iClk to the output on the Pin ( oDataB ) ;

so the clock-to-output time.

Note:

iClk is supposed to have 50Mhz, the PLL output is

also 50Mhz; the PLL is just used to compensate the

delay of the global clock network.

The essence (the part which I am worried about) is this:

For Cyclone IV devices TimeQuest reports a clock path delay

(from the input clock reference pin to an I/O register) of:

- fast corner: -0.298

- slow corner: 0.214

This is what I would expect:

The compensation keeps the delay close to 0ns. For the fast corner

you get a negative "delay", because the PLL over-compensates.

For Cyclone V devices TimeQuest reports the same kind of clock path delay as:

- fast corner: 0.591

- slow corner: 3.219

Especially the slow corner seems to indicate, that the PLL

is not compensating much.

Now I am wondering what am I doing wrong ?

Is this a known issue with Cyclone V devices ?