Forum Discussion
New ContributorIs "Cyclone 10 LP" slower than "Cyclone IV E" ?
I tested according to the same source code (VHDL) shown below.
The reported delay was almost the same...
but "Cyclone 10 LP" gave the following timing error @300MHz clock.
Is my setting incomplete? or Really "Cyclone 10 LP" slow?
---Timing Analyzer report @Cyclone 10 LP---
DEVICE:10CL006YU256C8G
; Slow 1200mV 85C Model Fmax Summary ;
+------------+-----------------+------------------------------------------------+---------------------------------------------------+
; Fmax ; Restricted Fmax ; Clock Name ; Note ;
+------------+-----------------+------------------------------------------------+---------------------------------------------------+
; 520.29 MHz ; 225.02 MHz ; pl|altpll_component|auto_generated|pll1|clk[0] ; limit due to minimum port rate restriction (tmin) ;
---
; Slow 1200mV 85C Model Minimum Pulse Width Summary ;
+------------------------------------------------+--------+---------------+
; Clock ; Slack ; End Point TNS ;
+------------------------------------------------+--------+---------------+
; pl|altpll_component|auto_generated|pll1|clk[0] ; -1.111 ; -1.111 ;
---Timing Analyzer report @Cyclone IV E---
DEVICE:EP4CE6F17C8
; Slow 1200mV 85C Model Fmax Summary ;
+------------+-----------------+------------------------------------------------+------------------------------------------------+
; Fmax ; Restricted Fmax ; Clock Name ; Note ;
+------------+-----------------+------------------------------------------------+------------------------------------------------+
; 520.29 MHz ; 402.09 MHz ; pl|altpll_component|auto_generated|pll1|clk[0] ; limit due to minimum period restriction (tmin) ;
---
; Slow 1200mV 85C Model Minimum Pulse Width Summary ;
+------------------------------------------------+-------+---------------+
; Clock ; Slack ; End Point TNS ;
+------------------------------------------------+-------+---------------+
; pl|altpll_component|auto_generated|pll1|clk[0] ; 0.846 ; 0.000 ;
---source code (vhdl)---
library ieee;
use ieee.std_logic_1164.all;
library altera_mf;
use altera_mf.altera_mf_components.all;
entity test is
port( rst, cki:in std_logic;
ti:in std_logic_vector(1 to 2);
do:out std_logic_vector(0 to 0));
end;
architecture a of test is
component altddio_out
generic(width:natural;
oe_reg:string:="UNUSED");
port( datain_h, datain_l:in std_logic_vector(width-1 downto 0);
outclock:in std_logic;
oe:in std_logic:='1';
dataout:out std_logic_vector(width-1 downto 0));
end component;
component plltst -- generated by Megawizard plugin manager
port( areset, inclk0:in std_logic:='0';
c0:out std_logic; -- 300MHz
locked:out std_logic);
end component;
signal c0:std_logic;
signal x:std_logic_vector(1 to 2);
begin
pl:plltst port map(rst, cki, c0);
ddo:altddio_out generic map(1)port map(x(1 to 1), x(2 to 2), c0, open, do);
process(c0)begin
if c0'event and c0='1' then
x<=ti;
end if;
end process;
end;
---compiler and other condition---
Quartus Prime Lite Edition 18.1.0.625
no user assignment ... so
I/O standard=2.5V, current strength=8mA, slew rate=2
11 Replies
KennyT_alteraSuper Contributor
each device had their own speed grade. Each of their own benefits and depends on your needs.
You can refer to https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/hb/cyclone-10/c10lp-51002.pdf
https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/hb/cyclone-v/cv_51002.pdf
Cyclone E usually is used for Optimized for lowest system cost and power for a wide spectrum of general logic and DSP applications.
Intel® Cyclone® 10 LP FPGAs usually used for optimized for low static power, low-cost applications, such as I/O expansion, sensor fusion, motor/motion control, chip-to-chip bridging, and control.
TNaka61Thankyou for fast responce ... but,
I made a supplementary test with a FF in an IO pad that is non-DDIO.
so, I found that the speed was virtually the same as shown below.
DEVICE:10CL006YU256C8G (Cyclone 10 LP)
; Slow 1200mV 85C Model Fmax Summary ;
+------------+-----------------+------------------------------------------------+------------------------------------------------+
; Fmax ; Restricted Fmax ; Clock Name ; Note ;
+------------+-----------------+------------------------------------------------+------------------------------------------------+
; 656.17 MHz ; 402.09 MHz ; pl|altpll_component|auto_generated|pll1|clk[0] ; limit due to minimum period restriction (tmin) ;
DEVICE:EP4CE6F17C8 (Cyclone IV E)
; Slow 1200mV 85C Model Fmax Summary ;
+-----------+-----------------+------------------------------------------------+------------------------------------------------+
; Fmax ; Restricted Fmax ; Clock Name ; Note ;
+-----------+-----------------+------------------------------------------------+------------------------------------------------+
; 656.6 MHz ; 402.09 MHz ; pl|altpll_component|auto_generated|pll1|clk[0] ; limit due to minimum period restriction (tmin) ;
From this result, it can be seen that only "Cyclone 10 LP"'s DDIO is about half the speed of non-DDIO,
and "Cyclone IV E"'s DDIO and non-DDIO are almost the same speed.
However, for both devices, the 0.5 cycle time is shorter than the Tco of DDIO output.
Therefore, the shortest pulse may disappear, but a long pulse should provide 0.5 cycle time resolution of the edge.
Is there a problem with my research method?
or Is either wrong?
- KennyT_altera
Can you check does your ddio and non ddio register is inside the IO?
Sometimes, quartus will not automatic put those register to the IO that would cause you to have a slower speed.
If the register is inside the IO, means your analysis is correct.
In arria 10, for speed that more than 200Mhz for DDIO. We would suggest customer to use phylite. But not for Cyclone 10LP.
- KennyT_altera
https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/ug/ug_altera_phylite.pdf
The IP core has a dedicated PHY clock tree in each I/O bank. The PHY clock tree is shorter which yields lower jitter and duty cycle distortion (DCD), enabling designs to achieve higher performance.
- TNaka61
>Can you check does your ddio and non ddio register is inside the IO?
I checked using the resource property editor on both devices.
This time it was necessary to specify "Fast output Register" explicitly on both devices
- KennyT_altera
Yes, my question is does your result above is base on the register reside inside the I/O?
- TNaka61
ok, My result above is based on a register placed inside an I / O block
and The result of the above placement was examined in the resource property editor
- KennyT_altera
If that is the case, then the comparison should be correct. As there are no better timing if those register that you have put inside the IO. Unless you use phylite or altdq_dqs2. Can you try to use AltDDIO to see if the result is different?
- KennyT_altera
I just recall somethings, can you go to timing analyzer and screenshot both delay out for comparison? I remember there are extra delay (like hmc delay stuff) and there is a workaround/avoid on this.
- TNaka61
Upload screen capture.
But I'm not sure if it's okay with this screen, so attach the project folder (except db, incremental_db).
The timing report for all cases is in the output_files folder of the attached file.
If there is no workaround, does it mean that cyclon10E can not be used instead of cyclon4E in this situation (like 300MHz x 2 time resolution)?
I'm very curious that cyclon10E's Tco is over 1ns faster under these conditions(2.5V,8mA,slew=2).
