 |
STAR SSD
|
- Choice of wafers for assembly on ladders
|
Pour la mise a jour du choix des batchs de ladders aller sur le site de la db:
Production>ladderBatch
The goal of this page is to present the way we may choose to arrange the 320 wafers into 20 ladders of 16 wafers each and to present an history of these choices since it may evolves with time. In other words : How to choose which wafers go where on which ladder ? The answer to this question may be given by an automatic procedure based on the reading of measurements contained in the SSD database and a 'clever' enough algorithm.
List of chosen detectors (NOT UP TO DATE)
Third sending to Thales - envoyes partiellement
(Results from choice 3 below)
- star_355, star_401, star_412, star_417, star_376, star_260, star_386, star_327, star_264, star_387, star_053, star_172, star_344, star_052, star_183, star_365, star_369, star_269, star_055, star_127 envoyes vers le 27 jan 2003
- star_322(broken),
- star_418, star_422, star_434, star_437, star_438, star_231, star_314, star_219, star_146, star_416, star_379, star_430, star_338, star_218, star_371, star_266, star_390, star_224, star_308, star_427 envoyes vers le 17 dec 2002
Second sending to Thales - 29 oct 2002
(Results from choice 3 below)
star_056, star_323, star_054, star_263, star_391, star_163, star_222, star_358, star_424, star_149, star_118, star_133, star_419, star_421, star_426, star_420, star_378, star_058, star_114, star_319
First sending to Thales - jul 2002
star_112, star_137, star_160, star_168, star_177, star_181, star_185, star_192, star_193, star_214, star_227, star_242, star_251, star_252, star_253, star_318, star_328, star_332, star_349, star_351, star_372, star_436
Updated choice - 27 March 2003
27 March 2003
L'algorithme equivalent au choix 3 du 26 juillet ci-dessous est re-tourne pour une mise a jour:
5 batchs de 72 wafers sont crees, voir le resultat.
Pour le batch 3 duquel on ete choisis les wafers envoyes a Thales, 13 wafers sont differents avec le choix du 26 juillet.
26 July 2002
Le 27 septembre 2002, le choix 3 (en rouge dans la table ci-dessous a ete choisi. Cliquez sur le lien 'succeed' puis sur [ladder constitution] pour obtenir la repartition des wafers dans les 5 lots correspondants.
La situation actuelle fait que les 20 detecteurs envoyes chez Thales au debut de l'annee (voir first choice) sont inutilisables (a quelques unites pres, peut-etre - voir Magalie). Sur les 415 detecteurs disponibles dans les qualites 1,2,3,6 et 4, en decomptant les 20 precites et les 20 utilises pour la premiere echelle (ladder 0) ils en restent 375 (355 si on enleve la qualite 4). C'est peu !. Surtout, s'il faut compter 10% de perte par Thales, nous n'aurions a la fin que 339=320+19 (ou 320 avec 4 qualites).
Sur ces bases, voici quelques propositions de choix.
| Choice |
1 |
2 |
3 |
4 |
5 |
6 |
| # batch |
5 |
5 |
5 |
7 |
7 |
7 |
| # wafers / batch |
71
= 4 ladders
+ 7 spares |
71
= 4 ladders
+ 7 spares |
72
= 4 ladders
+ 8 spares |
53
= 3 ladders
+ 5 spares |
53
= 3 ladders
+ 5 spares |
54
= 3 ladders
+ 6 spares |
| qualities |
1 2 3 6 |
1 2 3 6 |
1 2 3 4 6 |
1 2 3 6 |
1 2 3 6 |
1 2 3 4 6 |
| minimal voltage domain |
13 |
15 |
15 |
13 |
15 |
15 |
| ladders build |
20 |
19 |
20 |
20 |
19 |
20 |
| left wafers (unassociated) |
8 |
13 |
16 |
7 |
7 |
12 |
| full results |
succeed |
failed
but additional (bad) ladder possible |
succeed |
succeed |
failed
but additional (bad) ladder possible |
succeed |
First choice - end 2001
Apres la fabrication des premiers modules prototypes puis celle des modules qui ont ete montes sur l'echelle (20 en tout). Il a fallu choisir les wafers a envoyer a Thales pour le demarrage de la production.
Ce choix s'est fait suivant l'algorithme 3, dans la configuration 8x52, 5 qualities. 20 wafers on ete envoye chez Thales.
The input data:
Most of the 440 wafers (silicon detectors) prdeuced and tested matched the specification required and are ready to be assembled in the SSD. Nevertheless it is necessary to select among them those:
- with quality = 1, 2, 3 or 6 (high Vbias ones), maybe 4 (high currents)
- not assembled yet on a ladder,
- not used for module production tests by Thales.
This makes 386 for quality 1+2+3, 395 including quality 6 and 415 up to quality 4 wafers available, consult the production database for details.
What are we building ?
The SSD is made of 20 ladders supporting 16 modules each. We need some spare ladders, something like 3. Moreover we may consider to choose a little more than 16 modules to build a ladder. This will allow the replacement of an item accidentally broken in the assembly process. So parameters to keep in mind here are:
- 20+3 ladders,
- 16+1 modules per ladder.
Criteria to sort wafers:
First of all it has to be stated that it makes sense to select on the wafers to make ladders even if we don't have the module. Indeed it has been shown (see Magalie's work) that the main caracterisitcs of a wafer like depletion voltage or number of dead channels are almost identical for a wafer and a module.
So the characteristics on which we may choose the wafers are:
- quality (which insures the wafer is operating correctly and in reasonnable ranges with respect to voltages, leakage currents and number of dead strips),
- depletion voltage,
- breakdown voltage,
- operating range = breakdown - depletion voltage,
- total leakage current,
- number of dead strips on both sides.
What is a good ladder ?
A good ladder may be defined as something easy to operate (power up) and efficient which means as few dead channels as possible and as few noisy ones too. it has to be recalled here that the scheme of the biasing voltage distribution is such that all the wafers of the same are powered up at the same voltage. In this different respects we have to insure that :
- all wafers on a same ladder have a common voltage range of operation, the greater, the better,
- the chosen bias voltage is as near as possible to the depletion voltage of each wafer since we know this is the region where the noise is the lower,
- the summ of the leakage current stay below 200 uA (from CAEN specification),
- the overall number of dead channels is below few (1 ?) % for the entire ladder, and that they are randomly spread on it.
Another important point related to the assembly process is the question: "In which order do we assemble ladders ?"
For any algorithm described below, results will be showned the same way :
- list of wafers associated with ladders,
- diagrams of operating range of each ladder,
- diagrams of number of dead strips per ladder,
- diagrams of leakage current per ladder,
- for each ladder, diagrams of distribution of operating range of each wafers.
If the procedure fails, it is obviously indicated. The number of wafers left unassociated is precised as well as their identification.
Random choice (algo 0):
Description
Just to have some reference and a first feeling of the difficulty, we have randomly gather wafers onto ladders, just insuring that there is a minimal common operating voltage for wafers of the same ladder.
Comments
The first conclusion one should make here is that it is possible to get what we want. That is 23 ladders with 17 modules.
Then the other good news is that even with a random choice, dead trips number (always below 100 for each sidewhich makes < .8 % of the channels of a ladder) is really small, The leakage current is as well not an issue, being always below 20 microAmps. Thus we conclude that a reasonable algorithm to sort wafers could simply not take into account these parameters (dead strips and leakage current).
Coming back to the voltage a ladder may be bias with, one should note that about half the ladders end up with a minimal biasing above 50 volts. This is bad for BNL security reasons. The other bad thing is that when looking at each ladder individualy, it is clear that the effective operating range of the ladder is randomly(not surprisingly) distributed inside the operating range of a given wafer. This means that we may not operate a wafer in optimal noise condition (specifically when the biasng voltage is near the higher limit).
Algorithm 1:
Description
This method build all ladder at the same time. It first attributes to a wafer a new quality, the best wafer being the one with the lowest depletion voltage, the largest operating domain, the lowest leakage current and the smallest number of dead strips (summed up over each side). To do so it sorts wafers according to each of these variables and attributes it a rank as follow:
- depletion voltage in ascending order : the lowest the rank, the lowest the voltage,
- operating range : the lowest the rank, the largest the range,
- leakage current : the lowest the rank, the lowest the current,
- number of dead strips : the lowest the rank, the smallest the number.
The new quality is actually a linear combination of the ranks of the wafer in this categories. Thus the ordering depends on the different weight applyed to each category. The depletion voltage and operating range get the highest weight.
After this step, each attributed with one of the first 23 best wafers. The second wafer of each ladder is in turn chosen in the remaining best wafers always trying to insure a minimal operating voltage for the ladder. And so on and so on up to the last wafer of each ladder.
Comments
The improvement compared to the random choice method is not really significant. One can see that there are still almost 50% of the ladder above 50 volts of minimal biasing. The improvement is actually in the distribution of the operating range of individual wafers of the same ladder, they do match better for some ladders offering a better chance to operate a lower noise.
Algorithm 2:
Description
Here, the choice is made ladder after ladder.
Wafers are sorting in ascending order of depletion voltage. The first available wafer with the lowest depletion voltage is taken as the first wafer of a given ladder. Then the next wafer to be chosen is the one that maximizes the operating voltage. And so on up to make a full ladder.
Then the procedure restarts with the next ladder.
Comments
So far this is the best algorithm for two reasons. There is a limited number of ladder which have to be operated above 50 volts. And when looking at the distribution of individual wafer operating range in a given ladder, they are almost identical. This means that all wafers of a same ladder will be operated to a bias voltage as near as possible to their own depletion voltage.
Algorithm 3:
Description
This method is based on algortihm 2 but trys to solve 2 issues.
The first one is that algo 2 does only try one choice for the first wafer of each ladder. Thus, if it happens that it is impossible to fullfill all criteria with other wafers having this one first, the algo fails. Algo 3 solves this in allowing ten iteration, that is ten different choice for the first wafer of each ladder, so as to try to avoid special case.
The second problem with algo 2, which is actually not specific to it but difficult to solve because of the previous limitation, is related to the time order in which the ladder are assembled. Since operation domain of ladders are strongly ordered (due to the order of the wafer with Vbias) if one starts with the lowest voltage ladder and goes on, it may happen that at some point many very good spare modules would never have been used whereas other worst ones have been assembled. To avoid this, we run the 3rd algorithm asking for 8 batch of 51 (or so) modules. In this way 3 ladders can be built out of one big batch allowing to share the spare modules, so that at the end of the assembly of thiese 3 ladders very few modules have not been used. In this scenario, one could start the assembly of ladders within any batch.
Comments
Best choice is probably 7 batch of 52 wafers, which makes 21 ladders + an additional ladder (this is called 8x52, 5 qualities just above). Note there are additional spare modules from the wafers not associated in the automatic algorithm.
Technical issue
Currently, all this algorithms have been implemented in a ROOT macro benefiting from the mySQL capabilities of ROOT. You may get the source code here. Try it and enjoy building your own SSD! You will find all the parameters at the beginning of the file.
It is foreseen to provide the same possibility directly from the database interface itself.
Ladder 0
Here is the description of the first ladder which is being assembled:
Modules : (no order)
mod_015 mod_026 mod_093 mod_050 mod_096 mod_097 mod_103 mod_105 mod_106 mod_111 mod_115 mod_132 mod_237 mod_280 mod_045 mod_046
Characteristics
Operation domain: 32-39 V, total leakage current = 19.000 uA, number od dead strip P-side=72 N-side=117
Note: The operation domain is defined by the maximum of all the depletion votltages and the minimum of all the breakdown voltages for the wafers constituting the ladder.
Wafer characteristics
| name |
depletion voltage (V) |
breakdown voltage (V) |
| star_015 |
19 |
49 |
| star_026 |
26 |
61 |
| star_093 |
20 |
57 |
| star_050 |
22 |
60 |
| star_096 |
22 |
56 |
| star_097 |
21 |
39 |
| star_103 |
14 |
48 |
| star_106 |
14 |
46 |
| star_111 |
18 |
52 |
| star_115 |
32 |
61 |
| star_132 |
26 |
58 |
| star_237 |
25 |
56 |
| star_280 |
15 |
47 |
| star_107 |
|
|
| star_108 |
|
|
| star_046 |
22 |
60 |
|
|
Jerome Baudot
Last modified: Fri Jul 26 14:57:07 CEST 2002