All the MOSFETs have been traced (0-2.5A @ 35˚C) and the segments where NNNNPPPP sets will be found are now with heinz1 for measurement under operating conditions (0-4A @65˚C) using Patrick's equipment.
I think there will be enough NNNNPPPP sets for everybody in the F5X Batch 1 GB. It should not be long and I think heinz1 will keep us updated on the process.
The matching rig and the FETs are only with him for a couple of days so I don't think he is there just yet
Cheers,
Nic
I think there will be enough NNNNPPPP sets for everybody in the F5X Batch 1 GB. It should not be long and I think heinz1 will keep us updated on the process.
The matching rig and the FETs are only with him for a couple of days so I don't think he is there just yet
Cheers,
Nic
If you don't need NP-matching there will be absolutely no problem.For the record...I'm still very keen to get my hands 6 pairs curve matched 0-2.5A 35c
Plenty of NN and PP pairs.
Yes - Zhou is a good and possibly the only alternative.
I have traced quite a few of his devices and he deliver very good matching perfectly in line with what he promises.
In my experience the 2SJ201 devices have a very regular behavior so I would not worry to much about these not being matched at operating current/temperature. However, the 2SK1530 have very variable Yfs and should be matched at operating conditions.
For those who have already bought NP-matched (not PP-matched only) devices from Zhou I am likely to be able to help in finding significantly better matched 2SK1530s for their 2SJ201s.
I will likely do this free of charge for people who are going to get matched sets from me, but this service may also (FETs permitting....) be extended to others.
More about this later.
Nic
THe allure of these parts is that they can be true matced pairs. IF only the first GB members are getting those pairs, what differentiates these from other fets. Will we be paying significantly less than NP matches since we will not have this level of matching. It seems reasonable that folks who receive the NP matches should pay the premium, as they will be receiving the premiuim devices. Not trying to cause trouble, but there was talk of considerable money for these pairs at beginning of thread, at that is OK for PN match, but I will not pay that for just 4n's and 4'ps that are matched along like devices.
Nobody will have to pay for anything they don't want to and the pricing has not yet been settled (difficult to do before we know the exact number of NNNNPPPP sets), so please stop speculating. The only thing which is certain is that F5X Batch 1 subscribers will be served first as this is why the GB was started.
You right Nic.....if you can get hold of him these days...i'm still waiting for replies to 2 PMs and correction of wrong components sent...exactly these you are considering
What about the "confirmed to the group buy" column on the s/sheet where a majority of people said no? Any credits there ?
.
I can't see that "a majority of people said no". Only a few have indicated "yes", but that is clearly not the same thing. Actually I am surprised to see that somebody confirmed their participation without knowing the costWhat about the "confirmed to the group buy" column on the s/sheet where a majority of people said no? Any credits there ?
There may be credits for genuine interest, position on list(s) etc. if the number of PN-mathed sets allow this. For example, in respect of everybody on the GB list, the number of NNNNPPPP-sets that will be made available to the F5X Batch1 members will be limited to the number of amps that these people actually posses.
Last edited:
Measuring at operating conditions
What is really unique about this group buy is the way that the MOSFETs are measured and the way the data are elaborated/utilized. Matching of MOSFETs for DIY’er is typically conducted at conditions that are far away from the reality under which they will be operating in the amplifier: they are typically not measured at operating temperature; far away from the current bias implemented in the amp; using a DC signal (fixed Vgs), and the transconductance (Yfs, or “gain”) of the DUT is completely ignored.
With the help of Patrick/XEN-audio we have here done it the “right way” considering all of the above issues.
Measurement procedure
(1) The MOSFET device under testing (DUT) is firmly mounted on a large massive heat-block kept at a constant temperature (65˚C) using a microcontroller.
(2) The DUT is “conditioned” exposing it to a Vgs AC signal resulting in an Id swing of 0-4A. The device is radiating heat just as it would be in the amp.
(3) Once stable, a trace of Vgs/Id data is collected and used for subsequent data analysis.
As far as reasonably possible this procedure copies the condition in which the MOSFET will find itself in a warmed up F5X playing a test tone at max volume.
Data elaboration
The collected dataset (a trace of about 2000 Vgs/Id readings) is fitted to a 6th order polynomial equation by non-linear regression and the Vgs (intercept) and Yfs (slope) values at bias (2A) derived from this.
Device matching
With a finite number of devices (550 off 2SK1530 and 600 off 2SJ201, in our case) the process of device matching is a process of making qualified compromises.
N + N -> NN and NN + NN -> NNNN matching:
the N-N and NN-NN matching is done using a small software app developed with the help of the informatics core where I work. This app does a iterative weighted pairing taking into account both Vgs and Yfs focused on quality rather than yield. The number of devices we have is good enough to allow this pairing without relevant compromises. The pairs and quads exceeding a fixed tolerance limit for Vgs (0.5%, about 11 mV) and Yfs (5%, about 40-50 mS) are rejected. These are maximum limits and the typical match is quite a bit better.
P-devices were matched the same way.
The above procedure generates NNNN and PPPP quads with excellent Vgs and Yfs match between phases and channels.
NNNN -> PPPP matching:
as previously posted the Vgs distribution of the MOSFETs we have at our disposition display a limited overlap between N and P devices. Implementing the same tight 0.5% tolerance Vgs limit used for NN and NNNN matching, the yield of NNNNPPPPP matches would be around half a dozen of sets….
These would have to be sold for > 1000$/set, just enough to cover the cost of the devices…..
Reducing the tolerance limit for NNNN-PPPP Vgs offset to 2.5% (about 55 mV) the yield of NNNNPPPP matches increases to 26 sets and we consider these 1st grade sets. They are as close to perfect as possible with the MOSFETs we have, considering the yield required to cover the Batch1 F5X amps.
The only “defect” of these sets is the slight N-P offset, which possible effect on distortion (in my limited understanding) will anyway be canceled by the balanced nature of the F5X.
A number (10 sets) of NNNNPPPP-matches came close to the above specifications, but failed the strict tolerance requirement for Yfs or N-P offset defined for the 1st grade sets. These sets have exactly the same quality of Vgs match for NN and NNNN (and PP/PPPP), but have specs for Yfs or N-P offset slightly out of the decided tolerance limits. These are very good sets and their “2nd grade” classification should absolutely not be taken as “bad”. They are still tope grade and very likely to be better than anything you already have in your hands or will be able to get elsewhere.
To give an impression of the quality of the 2SK1530/2SJ201 pairs we will be offering, I have attached a figure showing the traces of a representative NNNNPPPP-set with the 2SK1530 illustrated in blue and the 2SJ201 in red. The left panel show the traces and the right panel the quantification of the Vgs and Yfs match after normalization.
I will be back with more information tomorrow.
Cheers,
Nic
What is really unique about this group buy is the way that the MOSFETs are measured and the way the data are elaborated/utilized. Matching of MOSFETs for DIY’er is typically conducted at conditions that are far away from the reality under which they will be operating in the amplifier: they are typically not measured at operating temperature; far away from the current bias implemented in the amp; using a DC signal (fixed Vgs), and the transconductance (Yfs, or “gain”) of the DUT is completely ignored.
With the help of Patrick/XEN-audio we have here done it the “right way” considering all of the above issues.
Measurement procedure
(1) The MOSFET device under testing (DUT) is firmly mounted on a large massive heat-block kept at a constant temperature (65˚C) using a microcontroller.
(2) The DUT is “conditioned” exposing it to a Vgs AC signal resulting in an Id swing of 0-4A. The device is radiating heat just as it would be in the amp.
(3) Once stable, a trace of Vgs/Id data is collected and used for subsequent data analysis.
As far as reasonably possible this procedure copies the condition in which the MOSFET will find itself in a warmed up F5X playing a test tone at max volume.
Data elaboration
The collected dataset (a trace of about 2000 Vgs/Id readings) is fitted to a 6th order polynomial equation by non-linear regression and the Vgs (intercept) and Yfs (slope) values at bias (2A) derived from this.
Device matching
With a finite number of devices (550 off 2SK1530 and 600 off 2SJ201, in our case) the process of device matching is a process of making qualified compromises.
N + N -> NN and NN + NN -> NNNN matching:
the N-N and NN-NN matching is done using a small software app developed with the help of the informatics core where I work. This app does a iterative weighted pairing taking into account both Vgs and Yfs focused on quality rather than yield. The number of devices we have is good enough to allow this pairing without relevant compromises. The pairs and quads exceeding a fixed tolerance limit for Vgs (0.5%, about 11 mV) and Yfs (5%, about 40-50 mS) are rejected. These are maximum limits and the typical match is quite a bit better.
P-devices were matched the same way.
The above procedure generates NNNN and PPPP quads with excellent Vgs and Yfs match between phases and channels.
NNNN -> PPPP matching:
as previously posted the Vgs distribution of the MOSFETs we have at our disposition display a limited overlap between N and P devices. Implementing the same tight 0.5% tolerance Vgs limit used for NN and NNNN matching, the yield of NNNNPPPPP matches would be around half a dozen of sets….
These would have to be sold for > 1000$/set, just enough to cover the cost of the devices…..
Reducing the tolerance limit for NNNN-PPPP Vgs offset to 2.5% (about 55 mV) the yield of NNNNPPPP matches increases to 26 sets and we consider these 1st grade sets. They are as close to perfect as possible with the MOSFETs we have, considering the yield required to cover the Batch1 F5X amps.
The only “defect” of these sets is the slight N-P offset, which possible effect on distortion (in my limited understanding) will anyway be canceled by the balanced nature of the F5X.
A number (10 sets) of NNNNPPPP-matches came close to the above specifications, but failed the strict tolerance requirement for Yfs or N-P offset defined for the 1st grade sets. These sets have exactly the same quality of Vgs match for NN and NNNN (and PP/PPPP), but have specs for Yfs or N-P offset slightly out of the decided tolerance limits. These are very good sets and their “2nd grade” classification should absolutely not be taken as “bad”. They are still tope grade and very likely to be better than anything you already have in your hands or will be able to get elsewhere.
To give an impression of the quality of the 2SK1530/2SJ201 pairs we will be offering, I have attached a figure showing the traces of a representative NNNNPPPP-set with the 2SK1530 illustrated in blue and the 2SJ201 in red. The left panel show the traces and the right panel the quantification of the Vgs and Yfs match after normalization.
I will be back with more information tomorrow.
Cheers,
Nic
Attachments
Last edited:
Those of us primarily involved in this matching project (heinz1, nikon1975 and myself) have been collecting pieces for the F5X build for more than a year and as the chances of a 2SK1530/2SJ201 group buy appeared slim, all of us had previously purchased NP-matched devices from the known vendors around here. These devices were obviously entered into the pool and measured at operating conditions like all the rest.
Although the goal of this merger was to improve the sets we can of course use the data measured for these devices to understand what is the penalty of measuring at "non-operation" conditions.
Here I have attached the result for the nikon1975 NNNNPPPP-set.
Please note that we are not questioning the accuracy of the vendors data that we actually believe is quite good. We are just observing how the set would measure in the operating F5X.
Looking at the positive side the PPPP-quad in this set is very good.
Although the goal of this merger was to improve the sets we can of course use the data measured for these devices to understand what is the penalty of measuring at "non-operation" conditions.
Here I have attached the result for the nikon1975 NNNNPPPP-set.
Please note that we are not questioning the accuracy of the vendors data that we actually believe is quite good. We are just observing how the set would measure in the operating F5X.
Looking at the positive side the PPPP-quad in this set is very good.
Attachments
Last edited:
I have noticed in matching that some devices are more affected by changes in temperature than others. For my GB, the N channel devices tracked very well regardless of temp, but the P channel begins to seperate with higher temps. Granted they are still within 10mV at temperatures similar to these, but I have seen similar results. Have you guys tested fets that were matched at a singular point at operating temperature. I have found interesting things when comparing to that method vs a curve tracer type setup.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.