1
00:00:00,040 --> 00:00:08,120
Idaho is well known around the world for chip production but here at micron's

2
00:00:05,080 --> 00:00:10,320
memory R&D center of excellence in Boise

3
00:00:08,120 --> 00:00:16,560
it's a different kind of chip that gets all the attention behind me is over

4
00:00:13,719 --> 00:00:20,680
200,000 Square ft of Cutting Edge tilk development fabrication and testing

5
00:00:18,760 --> 00:00:25,320
facilities and thanks to micron's sponsorship I will be bringing you guys

6
00:00:23,000 --> 00:00:31,039
deep under the hood not just theoretically you know very lithography

7
00:00:28,039 --> 00:00:33,399
much soldering Etc a key condition of my

8
00:00:31,039 --> 00:00:38,960
visit today was that I would be allowed to actually use the equipment and build

9
00:00:36,079 --> 00:00:44,640
my own ddr5 memory kit for my next system upgrade and well I'm here aren't

10
00:00:41,800 --> 00:00:44,640
I so let's

11
00:00:50,480 --> 00:00:56,879
go oh my goodness this is a really big

12
00:00:54,239 --> 00:01:01,760
parking lot Boise main site as it's known already hosts a small army of

13
00:00:59,559 --> 00:01:07,360
Hightech workers nearly 9,000 and many more are coming as part

14
00:01:04,479 --> 00:01:12,280
of a $15 billion investment plan that will see up to

15
00:01:09,520 --> 00:01:18,280
600,000 Square ft of sustainable clean room manufacturing space brought online

16
00:01:14,439 --> 00:01:21,920
starting in 2025 but its raise on Detra

17
00:01:18,280 --> 00:01:25,200
today is completely different I'm about

18
00:01:21,920 --> 00:01:28,000
to enter Fab 4 which contains many of

19
00:01:25,200 --> 00:01:32,960
the same incredible etching polishing and implantation machines that we saw in

20
00:01:30,040 --> 00:01:38,159
our Intel Fab tour the difference is that not a single wafer that comes out

21
00:01:35,479 --> 00:01:44,520
of these multi-million dollar machines will ever touch the hands of a micron

22
00:01:40,759 --> 00:01:47,560
customer that's right this entire class

23
00:01:44,520 --> 00:01:50,119
100 clean room exists for the sole

24
00:01:47,560 --> 00:01:55,360
purpose of testing new production techniques that Micron can then deploy

25
00:01:52,439 --> 00:02:00,560
at their manufacturing centers in Taiwan Singapore Japan and Virginia they

26
00:01:58,360 --> 00:02:04,600
couldn't even tell me the name of some of the emerging memory technologies that

27
00:02:02,719 --> 00:02:09,360
they're validating in here but what I will tell you is that it's way more

28
00:02:06,880 --> 00:02:14,519
advanced than even the highdensity 1 beta node Dam and 232 layer nen that

29
00:02:13,040 --> 00:02:18,640
Micron is currently shipping to customers what they were able to tell me

30
00:02:17,040 --> 00:02:23,519
though is that as part of the development of their upcoming one gamma

31
00:02:21,200 --> 00:02:28,879
and one Delta nodes they're currently installing one of as sml's

32
00:02:25,920 --> 00:02:35,720
state-of-the-art extreme ultraviolet or euv lithog graphy systems and euv is

33
00:02:32,640 --> 00:02:39,440
absolutely wild this is worth more than

34
00:02:35,720 --> 00:02:42,080
my car my house my office and my life

35
00:02:39,440 --> 00:02:48,920
one of the challenges with euv is that the wavelength of the light is so small

36
00:02:45,159 --> 00:02:51,920
that it gets absorbed by glass so then

37
00:02:48,920 --> 00:02:54,560
with lenses out of the picture asml had

38
00:02:51,920 --> 00:03:00,319
to resort to a system of mirrors that are manufactured to literally Atomic

39
00:02:57,800 --> 00:03:05,319
Precision in order to direct the light through the mask they were giving me

40
00:03:02,400 --> 00:03:10,920
some fun speeds and feeds for this boy apparently it weighs over 180,000 tons

41
00:03:08,680 --> 00:03:16,000
and took 3 747s just to ship all the parts here for

42
00:03:14,239 --> 00:03:20,760
it to be assembled unbelievable now you might think then

43
00:03:19,120 --> 00:03:25,599
that with the right equipment manufacturing RAM is simple right I mean

44
00:03:23,480 --> 00:03:30,480
JC creates the specification and assuming you've got deep enough pockets

45
00:03:27,360 --> 00:03:33,040
for asml shiniest new toy you just

46
00:03:30,480 --> 00:03:37,400
follow the recipe right well I asked some Micron folks if that's how it

47
00:03:34,760 --> 00:03:37,400
worked

48
00:03:39,720 --> 00:03:45,760
and that's fair enough I mean even ignoring the challenge is inherent with

49
00:03:44,400 --> 00:03:50,080
building something that's never been built before a lot goes wrong when

50
00:03:48,040 --> 00:03:56,079
you're building nanometer scale features onto silicon and specialized tools are

51
00:03:53,079 --> 00:04:00,159
needed to diagnose these problems meet

52
00:03:56,079 --> 00:04:03,599
the UF 3000 wafer probing machine this

53
00:04:00,159 --> 00:04:05,720
thing is super cool our fop full of dyes

54
00:04:03,599 --> 00:04:12,480
comes in here gets sucked into the machine aligned via the notch on one

55
00:04:08,760 --> 00:04:15,920
side of the wafer then slid over here

56
00:04:12,480 --> 00:04:17,959
under one of these guys then through a

57
00:04:15,920 --> 00:04:24,639
combination of Machine Vision and manual input these little tiny wires are

58
00:04:22,040 --> 00:04:29,360
actually lined up with the individual dyes which can then be tested to ensure

59
00:04:27,120 --> 00:04:33,960
that they're functioning as expected spoil sper alert a lot of the time they

60
00:04:31,680 --> 00:04:39,199
don't from this screen I can pick any die I want anywhere on the wafer say

61
00:04:36,680 --> 00:04:45,479
that one and the Chuck will reposition so that my probes should be exactly

62
00:04:42,280 --> 00:04:48,240
above the contact points on that die we

63
00:04:45,479 --> 00:04:52,880
can actually check this by going to it's uh image display change is that right

64
00:04:50,320 --> 00:04:58,080
and there it is you can actually see we tested this one because there are ever

65
00:04:55,360 --> 00:05:02,600
so light contact points from the last time those tiny hairlike probes touched

66
00:05:01,199 --> 00:05:07,160
them that's all you need to make electrical contact with these things and

67
00:05:04,680 --> 00:05:11,280
find out just how well they're working once everything's aligned the engineer

68
00:05:09,039 --> 00:05:15,080
uses this station to run test patterns which can either reveal a fully working

69
00:05:13,000 --> 00:05:19,440
die or one that needs further investigation investigation that may not

70
00:05:17,520 --> 00:05:24,240
be able to be done automatically that's where manual probing comes in now to be

71
00:05:21,880 --> 00:05:28,680
clear if you're just running basic tests you would never set up at a station like

72
00:05:26,280 --> 00:05:32,560
this this is for deep investigations into what's going on with brand new

73
00:05:30,720 --> 00:05:36,960
silicon out of the Fab and there are a couple of key benefits to doing it this

74
00:05:34,280 --> 00:05:43,639
way one you don't tie up an auto prober for hours or days at a time and two

75
00:05:40,400 --> 00:05:46,639
we're actually able to probe the dye

76
00:05:43,639 --> 00:05:48,560
even more precisely than with the auto

77
00:05:46,639 --> 00:05:52,880
prober so we just need to look around actually can you guys show me one of the

78
00:05:50,039 --> 00:05:58,800
Pico probe points oh I got to zoom in I can't even see it uh so if we if we

79
00:05:55,680 --> 00:06:01,039
change the focus here holy crap so if we

80
00:05:58,800 --> 00:06:05,600
go back out and then you zoom in are you flipping kidding me you guys already saw

81
00:06:03,400 --> 00:06:10,680
how small those probe pads were these are Pico probe points these will tell us

82
00:06:08,599 --> 00:06:16,960
not just what's happening on the io of the D but what's happening inside the D

83
00:06:14,880 --> 00:06:21,639
all right you guys see this that's a Pico probe it looks like a piece of dust

84
00:06:20,039 --> 00:06:27,479
on the glass you're never going to see it Andrew you don't stand a chance my

85
00:06:24,919 --> 00:06:32,280
man wait so you guys are expecting me to land this it seems like you're setting

86
00:06:29,720 --> 00:06:36,479
up to fail was that enough they started moving movie Magic were moving over to

87
00:06:34,720 --> 00:06:40,639
one that they already landed for me cuz they were apparently expecting me to RAM

88
00:06:38,319 --> 00:06:45,560
the probe into the die which I did not do by the way anyway the point is this

89
00:06:43,479 --> 00:06:49,759
is super cool you can see this one's landed you can also see the Scribe marks

90
00:06:48,080 --> 00:06:55,520
that help with Machine Vision orientation and that will disappear when

91
00:06:52,479 --> 00:06:58,400
schwamp schwamp the space between the Dy

92
00:06:55,520 --> 00:07:02,720
is actually cut away it's extremely small okay my bud here is going to try

93
00:07:00,879 --> 00:07:07,440
and land the Pico probe what's really cool is you don't even have to hit probe

94
00:07:04,800 --> 00:07:12,800
points with pico probes you can actually hit individual wires if the wafer has

95
00:07:10,520 --> 00:07:17,840
not been passivated yet which is like a a protective layer that goes over top of

96
00:07:14,560 --> 00:07:23,680
it man it's amazing how big that probe

97
00:07:17,840 --> 00:07:26,680
looks relative to the D like what oh oh

98
00:07:23,680 --> 00:07:28,520
you missed yeah that's you have to get

99
00:07:26,680 --> 00:07:32,240
the angle just right so to kind of just bend on the top of it that's fine I

100
00:07:30,360 --> 00:07:35,560
think the people get it it's really hard now the key KN among you might have

101
00:07:33,599 --> 00:07:40,000
noticed that this wafer on the manual probe machine here has been cut in half

102
00:07:38,240 --> 00:07:45,240
that's for good reason when you're working on early Parts every wafer is

103
00:07:43,039 --> 00:07:50,039
precious and you want as many Engineers eyes on them as possible so by cutting

104
00:07:47,199 --> 00:07:55,759
them in half you double the number and so on and so forth until finally as a

105
00:07:53,319 --> 00:08:00,000
team you've managed to find and squash all the bugs yielding a working wafer

106
00:07:58,720 --> 00:08:04,319
from the Fab and here it is but before we can

107
00:08:02,639 --> 00:08:08,039
encapsulate these dieses and build them into working memory modules there are a

108
00:08:06,280 --> 00:08:12,800
couple more things we need to do starting with thinning you see this is

109
00:08:10,479 --> 00:08:17,000
too thick to build into a working memory die especially if you're stacking them

110
00:08:14,800 --> 00:08:21,919
on top of each other something Micron is intimately familiar with so our first

111
00:08:19,759 --> 00:08:27,199
step is to put it in this machine where a thin plastic protective film will be

112
00:08:24,520 --> 00:08:32,479
applied to the entire wafer after that our wafer gets flipped upside down on

113
00:08:29,520 --> 00:08:36,240
into that plastic layer and goes into the

114
00:08:33,320 --> 00:08:41,919
grinder as advertised it grinds our wafer down to a thickness Micron would

115
00:08:38,839 --> 00:08:45,000
not specify for me but suffice it to say

116
00:08:41,919 --> 00:08:47,800
a lot thinner and applies it to a blue

117
00:08:45,000 --> 00:08:52,920
sticky back piece of plastic just like this one and the reasons for this are

118
00:08:49,880 --> 00:08:54,760
twofold first of all is the fact that

119
00:08:52,920 --> 00:09:01,720
thinned Wafers tend to do what they call in Idaho Potato chipping where they curl

120
00:08:58,320 --> 00:09:04,800
up like this and number two is that they

121
00:09:01,720 --> 00:09:08,480
need a way to hold all of the individual

122
00:09:04,800 --> 00:09:10,959
rectangular dyes in place during and

123
00:09:08,480 --> 00:09:17,720
after the slicing process that takes place in here with saws that have

124
00:09:13,880 --> 00:09:19,519
Diamond blades that are literally the

125
00:09:17,720 --> 00:09:24,079
width of a human hair and I'm not talking a head hair I'm talking one of

126
00:09:21,600 --> 00:09:28,240
the tiny little wimpy end of your finger hairs probably the most impressive thing

127
00:09:26,040 --> 00:09:32,920
about it is how precisely controlled it is not only does it cut all of our

128
00:09:30,279 --> 00:09:38,279
dieses perfect on the X and Y AIS it cuts to a perfect depth such that the

129
00:09:35,480 --> 00:09:42,959
dieses are fully cut and yet still adhered to our sticky back once we reach

130
00:09:41,000 --> 00:09:48,200
this stage we're ready for encapsulation or packaging now obviously we can't even

131
00:09:46,240 --> 00:09:52,440
see some of the features on these dyes with the naked eye let alone solder a

132
00:09:50,480 --> 00:09:56,079
wire to them so packaging is a whole other branch of black magic unto itself

133
00:09:54,880 --> 00:10:01,320
they don't do any of it here unfortunately but in a nutshell for a 3D

134
00:09:58,800 --> 00:10:05,920
product you would stack these dyes thermally Bond them with Incredible

135
00:10:03,680 --> 00:10:09,680
Precision attach these Bond wires so that you can actually solder the bloody

136
00:10:07,440 --> 00:10:14,720
thing to aboard then embed the whole thing in an epoxy or mold compound so

137
00:10:12,320 --> 00:10:18,760
that it stays safe not just in transit but also during use you can see this one

138
00:10:17,040 --> 00:10:23,959
right here has actually had that mold compound sanded away so that you can see

139
00:10:21,120 --> 00:10:28,160
the be dye material but this familiar plain black rectangle is what's actually

140
00:10:26,079 --> 00:10:33,320
going to make its way to the next stage the characterization tester now up until

141
00:10:31,160 --> 00:10:37,680
now we've gone ahead and we've probed these dies to ensure that they do work

142
00:10:35,519 --> 00:10:42,000
as intended but that's just for show after encapsulation we should be left

143
00:10:39,839 --> 00:10:45,560
with the familiar black square that looks just like what you'd find on a

144
00:10:43,560 --> 00:10:49,800
finished memory module except that all the testing we've done on it up until

145
00:10:47,120 --> 00:10:54,399
now is not sufficient you see the issue with our probes is that their long leads

146
00:10:52,720 --> 00:11:00,000
are susceptible to signaling interference and cross talk which means

147
00:10:57,240 --> 00:11:04,800
that they need to be run at a much much lower speed than what micron's customers

148
00:11:02,639 --> 00:11:09,519
demand so while we know that these dies work in theory we need to plug them into

149
00:11:06,959 --> 00:11:13,600
one of these to see if they still work at speed I'm not going to put it on this

150
00:11:11,639 --> 00:11:18,240
one do you guys have another one they dug me up a ddr4 load board that I can't

151
00:11:15,959 --> 00:11:23,040
do too much damage to at this point they also gave me a ddr4 die basically the

152
00:11:20,880 --> 00:11:29,079
purpose of this board is to make it much much easier to probe all the different

153
00:11:25,639 --> 00:11:30,680
pins on the bottom of our packaged di so

154
00:11:29,079 --> 00:11:36,200
we put that little guy in there like that then this hold down mechanism

155
00:11:34,320 --> 00:11:40,600
clamps it onto the board so that we don't have to solder it and throw these

156
00:11:38,399 --> 00:11:45,600
away every time or or desolder them or whatever the case may be the load board

157
00:11:43,360 --> 00:11:50,079
then goes onto our testing table here and man I'm going to need an explanation

158
00:11:48,200 --> 00:11:54,240
of exactly what it is that is going on with our Scopes here there's obviously a

159
00:11:51,959 --> 00:11:58,839
lot more to it but the basics are that our yellow line is our data strobe our

160
00:11:56,600 --> 00:12:03,760
blue is our clock and our green and red lines our input and output and what we

161
00:12:01,399 --> 00:12:09,320
want to see is we want to see these lines as vertical as possible I mean a

162
00:12:06,720 --> 00:12:14,160
theoretical perfect Digital Signal would actually have flat Peaks and valleys and

163
00:12:12,360 --> 00:12:19,160
straight lines between them but in the real world it takes a little bit of time

164
00:12:16,519 --> 00:12:22,680
to go from your low Target voltage to your high Target voltage especially when

165
00:12:21,160 --> 00:12:27,560
you're operating at speeds in excess of a billion cycles per second another fun

166
00:12:25,079 --> 00:12:32,279
bit is the kind of oscilloscope that you need in order to run this kind of

167
00:12:29,279 --> 00:12:35,320
testing is uh worth a pretty penny this

168
00:12:32,279 --> 00:12:37,360
bad boy apparently do 33 GHz now

169
00:12:35,320 --> 00:12:42,160
obviously micron's not going to out their suppliers for their exact pricing

170
00:12:39,720 --> 00:12:46,560
but I've been told that this scope with probes and software and all the other

171
00:12:44,000 --> 00:12:51,639
bits and Bobs you might need could be in the neighborhood of a half a million

172
00:12:48,480 --> 00:12:53,560
dollars so to be clear this is not the

173
00:12:51,639 --> 00:12:59,959
kind of thing that they're running at mass production this is an R&D process

174
00:12:57,000 --> 00:13:04,440
where you know what this kind of C honestly compared to the equipment it

175
00:13:01,639 --> 00:13:10,160
took to make that one decent wafer with a couple usable dieses absolutely dwarfs

176
00:13:07,600 --> 00:13:15,240
this of course not every issue can be diagnosed with probing that is where

177
00:13:13,040 --> 00:13:20,760
this x-ray machine comes in whether we're looking at an individual die or a

178
00:13:17,519 --> 00:13:22,880
full module this gives us a 3D

179
00:13:20,760 --> 00:13:27,720
non-destructive look at where our problem might lie all we have to do is

180
00:13:25,279 --> 00:13:32,760
load our sample onto this carrier pop it in here black last our sample with

181
00:13:30,199 --> 00:13:38,320
x-rays which are captured on this side while this pedestal slowly rotates

182
00:13:36,399 --> 00:13:42,959
generating over here this is made of lead so it's very heavy I'm just going

183
00:13:40,040 --> 00:13:50,040
to be careful closing that generating this 3D model that's a composite of all

184
00:13:47,199 --> 00:13:55,040
of the individual scans so this is the PCB under here this right here is a pic

185
00:13:52,759 --> 00:13:59,720
module and this right here is our solder joint and if we look closely you can see

186
00:13:57,680 --> 00:14:05,399
the whole thing cracked apart during a stress test that's no good what's really

187
00:14:03,279 --> 00:14:10,240
wild though is that's one of the lowest resolution Tools in this lab behind this

188
00:14:08,199 --> 00:14:14,040
door which unfortunately I'm not allowed to open are scanning electron

189
00:14:11,959 --> 00:14:19,920
microscopes that can go down to the gate and transistor level helping to diagnose

190
00:14:16,920 --> 00:14:23,199
any problems in the dyes until finally

191
00:14:19,920 --> 00:14:26,399
we get maybe one or two good Wafers at

192
00:14:23,199 --> 00:14:28,160
which stage we move to our next station

193
00:14:26,399 --> 00:14:34,959
but while the characterization tester tells us what an experimental design may

194
00:14:31,120 --> 00:14:36,880
handle the ddr6 6000 something who knows

195
00:14:34,959 --> 00:14:41,519
that doesn't mean that it's ready for prime time they might have just gotten

196
00:14:38,759 --> 00:14:46,959
lucky so the next step is to go back to the Fab and say let's run a larger batch

197
00:14:44,560 --> 00:14:51,880
maybe a 100 W first this time encapsulate those and send them over

198
00:14:49,480 --> 00:14:56,839
here this is a bining machine and it helps micron's Engineers figure out what

199
00:14:54,199 --> 00:15:01,360
the variation is with a much larger sample size I mean maybe

200
00:14:59,560 --> 00:15:06,759
absolutely everything that comes in here runs on target or above but much more

201
00:15:04,759 --> 00:15:13,320
likely there's going to be some failed dies or other Corner cases so this

202
00:15:09,800 --> 00:15:15,680
machine takes in trays of packaged dyes

203
00:15:13,320 --> 00:15:21,279
and puts them through hell intensive operations hot cold the works then the

204
00:15:19,639 --> 00:15:26,320
dyes get sorted according to the engineer's parameters in the outputs

205
00:15:23,880 --> 00:15:31,560
down here some of these problems might be fixable ones you know maybe this spin

206
00:15:29,000 --> 00:15:37,199
could be sold as a slower part or maybe this bin has fuses that you could blow

207
00:15:33,920 --> 00:15:39,319
so you got a 16 GB die and you know half

208
00:15:37,199 --> 00:15:44,680
of it's fine but you just disable the other half and now it's an 8 gigabit die

209
00:15:42,040 --> 00:15:50,040
that stuff happens all the time but when it does that's not a prime dye anymore

210
00:15:47,720 --> 00:15:55,720
the dyes that go into micron's consumer crucial branded dims those are all going

211
00:15:52,560 --> 00:15:58,160
to be prime D dies and on that subject I

212
00:15:55,720 --> 00:16:03,800
think we've actually got some here so it might be time for to go build my modules

213
00:16:01,440 --> 00:16:08,079
now that we have trays upon trays of these packaged dieses that will run

214
00:16:05,800 --> 00:16:13,480
perfectly at their rated characteristics I am so excited right now it is time for

215
00:16:10,440 --> 00:16:17,440
us to build some actual

216
00:16:13,480 --> 00:16:19,920
modules uh not like that for that we're

217
00:16:17,440 --> 00:16:24,480
going to need the machine behind me this is a multi-stage automated PCB

218
00:16:22,399 --> 00:16:28,000
production line very similar to what crucial might use to build the server

219
00:16:26,160 --> 00:16:31,959
memory modules that are powering the machine that you are watching watching

220
00:16:29,199 --> 00:16:34,880
this video from right now the first stage is silk screening which is pretty

221
00:16:33,440 --> 00:16:39,720
much the same process that we use to print our t-shirts on LTT Store.com

222
00:16:37,440 --> 00:16:46,040
except instead of shirts we use what I'm informed is called a panel of memory

223
00:16:42,120 --> 00:16:48,440
module pcbs and okay actually this

224
00:16:46,040 --> 00:16:52,680
part's pretty much the same so we pop our panel down here an electronic eye

225
00:16:50,680 --> 00:16:57,519
detects that it's correctly placed then my lovely assistant over here presses

226
00:16:54,480 --> 00:16:59,560
the button normally this machine would

227
00:16:57,519 --> 00:17:04,120
not operate with the cover open like this and I have been cautioned to keep

228
00:17:01,560 --> 00:17:08,520
my hands well away from it which I will do oh there it goes wow that's really

229
00:17:06,559 --> 00:17:15,480
fast so you can see that there's kind of a little squeegee that's wiping solder

230
00:17:12,160 --> 00:17:17,679
paste Sho it's like already done wait is

231
00:17:15,480 --> 00:17:20,199
it already coming out the other side oh my God it's going all the way into the

232
00:17:19,000 --> 00:17:24,720
other machine already we didn't even get a chance to look at it fortunately I

233
00:17:22,000 --> 00:17:31,559
know what's happening here this is oh not good we got a defect this is so cool

234
00:17:28,640 --> 00:17:37,720
I'm so glad we got a defect this machine creates a 3D computer vision model of

235
00:17:34,919 --> 00:17:43,000
our panel of memory sticks and you can see here it can actually detect

236
00:17:40,000 --> 00:17:45,600
everything from uh two contacts that are

237
00:17:43,000 --> 00:17:49,720
bridged like right here that would be a big problem when we soldered our chips

238
00:17:47,320 --> 00:17:53,640
onto it to ones that actually have too much or too little solder paste applied

239
00:17:51,799 --> 00:17:56,799
to them it looks like all of the errors we got were bridging errors which means

240
00:17:55,360 --> 00:18:02,480
we're going to have to try again I guess or can we go in and fix it we can fix it

241
00:18:00,480 --> 00:18:05,520
this is more fun than I expected I've got permission to just open this up

242
00:18:04,080 --> 00:18:11,880
we're just going to we're just going to take our soon to modules out okay so

243
00:18:09,520 --> 00:18:17,200
where's our is this the thing we're looking at here must be I'm just

244
00:18:14,919 --> 00:18:20,960
figuring this out in real time here oh yeah I see it and given that this is

245
00:18:18,919 --> 00:18:24,799
part of the power management of the dim I pretty much promise you this thing

246
00:18:22,919 --> 00:18:30,120
lights on fire and lets out the Magic Smoke the second you plug it in uh okay

247
00:18:27,440 --> 00:18:34,679
so what do I do now unfortunately I can't show you guys exactly what's going

248
00:18:33,240 --> 00:18:39,559
on here oh man am I even going to be able to

249
00:18:36,559 --> 00:18:41,480
hold okay this is uh this is some

250
00:18:39,559 --> 00:18:47,679
Precision work and I can't even rest my hand on the rest of the panel

251
00:18:43,880 --> 00:18:49,280
so okay I got one of them I think now

252
00:18:47,679 --> 00:18:53,320
you would never do this in mass production would you is this just for

253
00:18:51,080 --> 00:18:58,000
like test batches thank goodness now that I think

254
00:18:56,240 --> 00:19:02,159
I fixed the problem it's entirely possible that AI didn't and B I screwed

255
00:19:00,559 --> 00:19:08,240
up something else so I'm going to go ahead and load this back on here let's

256
00:19:05,280 --> 00:19:13,760
see if we nailed it here yes yes yes yes yes yes yes yes yes yes not good wait I

257
00:19:12,080 --> 00:19:17,280
never actually fixed the ones that were the defects last time I fixed something

258
00:19:15,440 --> 00:19:21,360
else they apparently thought it was Charming how much fun I was having so

259
00:19:19,000 --> 00:19:26,559
they didn't stop me but this board is bad because of that that is not

260
00:19:24,080 --> 00:19:29,760
something I can fix with an xacto knife time for round two I asked about the

261
00:19:28,400 --> 00:19:33,960
solder water paste they're using by the way even this stuff is cool it comes out

262
00:19:31,799 --> 00:19:39,360
of a refrigerated tube that is like surprisingly cold and really really

263
00:19:37,600 --> 00:19:44,799
heavy they won't tell me exactly what it's made of but we know for sure it

264
00:19:41,600 --> 00:19:47,400
isn't lead so I'm fairly certain you can

265
00:19:44,799 --> 00:19:54,080
eat it without dying or at least without dying of lead poisoning anyway 76 errors

266
00:19:51,000 --> 00:19:55,960
this time oh boy actually you know what

267
00:19:54,080 --> 00:20:01,280
this might not be that bad most of these look like they say insufficient solder

268
00:19:58,760 --> 00:20:05,840
it's pretty clear they're probably fine I mean can I just try like can I try it

269
00:20:04,520 --> 00:20:11,840
as long as there's no Bridges okay so if I fix the bridges then we can go for it

270
00:20:09,000 --> 00:20:15,280
I mean the worst case scenario is I make my own defective memory right like it's

271
00:20:13,799 --> 00:20:19,320
not like you guys are shipping it to anyone final attempt and we're going

272
00:20:17,440 --> 00:20:26,720
forward no matter what happens here what do you think we got this all right let's

273
00:20:22,200 --> 00:20:28,880
try it I okay so right I uh I click pass

274
00:20:26,720 --> 00:20:34,280
okay so I have overridden the machine that machine next up is our assembly

275
00:20:31,480 --> 00:20:38,400
robots these helpful guys pick surface mount components off of these reels that

276
00:20:37,120 --> 00:20:43,080
are loaded into the bottom of the machine here and then depending on the

277
00:20:41,039 --> 00:20:47,039
type of component they have different heads that you can put on them so this

278
00:20:44,679 --> 00:20:52,240
is obviously for a very small component whereas this is for a large one and the

279
00:20:49,640 --> 00:20:56,720
robot knows what kind of head it has using machine vision and reading these

280
00:20:54,640 --> 00:21:03,080
dots on the back so it's got a little vacuum that allows it to pick up one

281
00:20:59,799 --> 00:21:05,159
single part and then place it onto our

282
00:21:03,080 --> 00:21:09,400
memory modules that's why they're more formally known as pick and place

283
00:21:07,000 --> 00:21:12,679
machines although I like my name too that process with the tape loaded

284
00:21:10,600 --> 00:21:17,240
components actually takes a few minutes because there are so many of them then

285
00:21:15,480 --> 00:21:21,200
when it's done we're going to pull instead from a tray of packaged dieses

286
00:21:19,559 --> 00:21:25,480
which is loaded in over here and then the real good stuff goes on now we're

287
00:21:23,520 --> 00:21:31,200
ready for the last step which is basically a fancy pizza oven there it

288
00:21:29,000 --> 00:21:35,640
goes it's designed to bring our panel up to exactly the perfect temperature hold

289
00:21:33,520 --> 00:21:41,480
it there for a while then drop it back down heat it back up and cool it down

290
00:21:39,159 --> 00:21:46,120
now this isn't one of the fancy gravity defying ones that allows you to pick and

291
00:21:43,679 --> 00:21:50,039
place Parts on both sides using kind of like a tacky solder paste uh you'd be

292
00:21:48,600 --> 00:21:55,480
more likely to find something like that on a mass production line but the test

293
00:21:53,080 --> 00:21:59,039
Engineers have a pretty cool solution to their single-sided oven they can

294
00:21:57,880 --> 00:22:05,720
actually prep ahead of time dozens or wow actually

295
00:22:02,279 --> 00:22:08,159
like even hundreds of pcbs that have all

296
00:22:05,720 --> 00:22:11,640
the little picky components on one side and then the other side's blank so all

297
00:22:10,120 --> 00:22:16,080
they have to do is pick in place the eight packaged dyes throw them through

298
00:22:13,840 --> 00:22:22,080
the oven and they're ready to test this is it it's cooked but um obviously

299
00:22:20,400 --> 00:22:26,760
putting this into a computer is going to be a bit of a problem so they have a

300
00:22:24,000 --> 00:22:30,400
solution for that it's this little CNC cutter that I have not been briefed on

301
00:22:28,600 --> 00:22:35,000
how to use yet but it couldn't be that hard right I mean everything so far has

302
00:22:32,520 --> 00:22:41,320
gone in this way so surely it goes in head first and then oh well hopefully I

303
00:22:37,840 --> 00:22:42,919
put it in the right way okay so start

304
00:22:41,320 --> 00:22:46,200
I'm getting surprisingly little supervision here so I'm just doing it

305
00:22:44,720 --> 00:22:50,880
here we go oh this thing's sick Get Over Here Andrew okay so it goes what it goes

306
00:22:48,880 --> 00:22:54,799
and gets the right cutting tool or what are we looking at here it's a board

307
00:22:52,880 --> 00:22:59,000
gripper ah so it's got to get the right board gripper you might be doing sood

308
00:22:56,400 --> 00:23:03,200
dims or dims or ddr4 ddr5 okay here comes my module oh am I

309
00:23:02,360 --> 00:23:07,520
too close that error was caused by one of

310
00:23:05,600 --> 00:23:11,640
the safety checks that prevents the wrong profile from running for the kind

311
00:23:09,679 --> 00:23:15,279
of memory that you have so it actually reads the type of modules that are

312
00:23:13,679 --> 00:23:19,080
coming in and if it's got the wrong gripper bit loaded it's going to say hey

313
00:23:16,919 --> 00:23:22,240
whoa whoa whoa you can't do that but it's really cool that we got that error

314
00:23:20,720 --> 00:23:27,720
because it gave us an excuse to open up this machine and take a closer look at

315
00:23:24,120 --> 00:23:29,520
it from the top it's just a gripper hand

316
00:23:27,720 --> 00:23:33,919
from the bottom there's actually a router head that cuts off all the extra

317
00:23:32,120 --> 00:23:37,440
little tabs and then the gripper is going to take the finished modules and

318
00:23:35,840 --> 00:23:40,760
move them over to that conveyor at the back okay you guys can go ahead and

319
00:23:38,880 --> 00:23:45,000
close it let's run it this is cool by the way remember that video that we did

320
00:23:42,880 --> 00:23:49,120
on gold recovery from Electronics manufacturing back in Taiwan well this

321
00:23:47,120 --> 00:23:53,600
is exactly the kind of stuff that would get shipped to them to be broken down

322
00:23:51,320 --> 00:23:59,919
and turned back into usable materials here they are and these are

323
00:23:56,200 --> 00:24:02,320
ready for my computer at home no oh

324
00:23:59,919 --> 00:24:07,600
right I don't even know what I've built yet but before I can find out we've got

325
00:24:04,640 --> 00:24:12,520
one more really important stopover the long-term Burnin and validation lab they

326
00:24:10,640 --> 00:24:17,240
are constantly building out more capacity because you know what

327
00:24:14,480 --> 00:24:22,039
theoretically works is not good enough so Micron is constantly adding more real

328
00:24:19,799 --> 00:24:27,600
servers the exact kind that their customers might put their memory in and

329
00:24:24,440 --> 00:24:31,240
then just absolutely pummeling these

330
00:24:27,600 --> 00:24:34,080
things for talking hundreds thousands

331
00:24:31,240 --> 00:24:39,799
upon thousands of hours to see what exactly could possibly go wrong ideally

332
00:24:36,919 --> 00:24:43,320
nothing knock on metal I'm not allowed to tell you how many server racks they

333
00:24:41,520 --> 00:24:49,080
have in here but let's just put it this way there's a lot of electrical panels

334
00:24:46,200 --> 00:24:53,640
and they're testing a lot more than just current gen and the same goes for

335
00:24:51,399 --> 00:24:59,799
Consumer motherboards on the wall of fame here they've got validation boards

336
00:24:56,919 --> 00:25:05,360
for basically every platform that has ever existed so that even if years down

337
00:25:03,320 --> 00:25:10,760
the line a customer complaint comes through they can go back to the board

338
00:25:07,399 --> 00:25:13,080
maker the chipset and CPU maker and work

339
00:25:10,760 --> 00:25:18,200
together to figure out who's causing the problem and what the solution might be

340
00:25:16,480 --> 00:25:22,480
what really blew my mind in this room though wasn't the environmental Chambers

341
00:25:20,440 --> 00:25:26,159
or the collection of old motherboards but rather the collection of weird

342
00:25:24,720 --> 00:25:30,480
motherboards the likes of which I'd never seen like the thing is you think

343
00:25:28,720 --> 00:25:35,960
Micron they probably got to collaborate with Brands you think of like an Intel

344
00:25:32,840 --> 00:25:38,559
or an AMD or an apple but these days

345
00:25:35,960 --> 00:25:43,159
pretty much everything has a chip in it and every chip needs RAM so they also

346
00:25:41,120 --> 00:25:48,440
work with Brands you might not think of Automotive companies dishwasher

347
00:25:45,600 --> 00:25:52,520
companies you name it it's probably got Micron Hardware in it and so there's

348
00:25:50,480 --> 00:25:57,200
wild stuff in here like this thing we can't show it to you obviously but it's

349
00:25:54,799 --> 00:26:03,799
three times the size of a standard ATX motherboard and absolutely covered in IO

350
00:26:00,960 --> 00:26:10,399
that I have never even heard of what even is it just because it's functional

351
00:26:07,200 --> 00:26:13,320
doesn't mean it's compliant it might

352
00:26:10,399 --> 00:26:19,320
seem like Micron is just flexing their sick 3M X 3M anoa chamber on me cuz they

353
00:26:16,919 --> 00:26:23,279
know mine is broken but no this is an incredibly important part of the

354
00:26:20,880 --> 00:26:28,600
development process so we load our system or our module or our SSD or

355
00:26:26,919 --> 00:26:33,919
whatever the case may be on to this turntable load up an exerciser which is

356
00:26:31,399 --> 00:26:39,200
a program that stresses the component and the turntable spins around with this

357
00:26:36,159 --> 00:26:41,440
antenna here detecting the highest

358
00:26:39,200 --> 00:26:45,919
emissions off the product from every direction this ensures that it meets

359
00:26:43,320 --> 00:26:50,520
regulatory requirements and perhaps more importantly ensures that it doesn't spew

360
00:26:48,200 --> 00:26:54,360
radiation that could cause bit flips or other Gremlins to pop up in a given

361
00:26:52,279 --> 00:27:00,000
system oh on that subject by the way they also intentionally bombard their

362
00:26:57,039 --> 00:27:04,760
parts with em Mii while monitoring for abnormal behavior or instability either

363
00:27:02,200 --> 00:27:09,039
of which would of course be unacceptable that is one hell of an antenna you're

364
00:27:06,640 --> 00:27:14,840
compensating for something hilariously it was a real challenge to find anywhere

365
00:27:11,440 --> 00:27:17,640
to test my modules see the thing is ddr5

366
00:27:14,840 --> 00:27:21,760
is very current and so all the ddr5 capable gear is busy having the crap

367
00:27:19,799 --> 00:27:25,679
kicked out of it in one of micron's testing Labs that is chalk full of

368
00:27:24,039 --> 00:27:29,960
unreleased Hardware so we weren't going to be able to film there fortunately we

369
00:27:27,640 --> 00:27:33,399
found a single test bench and a quiet Corner that hopefully has working power

370
00:27:32,200 --> 00:27:37,919
where we are going to hook up these modules and see if we can actually post

371
00:27:35,799 --> 00:27:41,840
this thing This Is Amazing by the way people just like set up a test bench for

372
00:27:39,679 --> 00:27:46,159
you I have to set up my own test benches back at the office it's

373
00:27:43,320 --> 00:27:51,640
ridiculous I still have no idea what we built but they did tell me they had

374
00:27:48,159 --> 00:27:54,679
these D flown in specially so maybe

375
00:27:51,640 --> 00:27:56,799
there's something pretty cool

376
00:27:54,679 --> 00:28:01,720
A2 hey we got something CPU or memory

377
00:27:59,720 --> 00:28:07,720
changed look at that we went straight for gold and we got it did you guys test

378
00:28:03,399 --> 00:28:10,799
these ahead of time no we just went went

379
00:28:07,720 --> 00:28:15,159
for it oh wow you guys are ballsy I like

380
00:28:10,799 --> 00:28:18,880
it uh-oh we have 96 G wait what the hell

381
00:28:15,159 --> 00:28:21,320
am I reading this right these are 24 gab

382
00:28:18,880 --> 00:28:26,760
modules no one's volunteering what the hell is a 24 gig module explain yourself

383
00:28:25,480 --> 00:28:32,960
uh I can't that's why nobody volunteered okay

384
00:28:29,640 --> 00:28:36,600
so how did what what now this is obvious

385
00:28:32,960 --> 00:28:39,679
1337 for the part number 8

386
00:28:36,600 --> 00:28:40,799
terabytes DDR 9 9,000 that's a missed

387
00:28:39,679 --> 00:28:44,919
opportunity it should have been over 9,000 obviously okay but for but for

388
00:28:43,440 --> 00:28:49,519
real you got to you got to tell you got to give me something to work with here

389
00:28:46,519 --> 00:28:49,519
so

390
00:28:59,679 --> 00:29:06,760
and that is why we had to bleep all of that I really am not allowed to explain

391
00:29:04,200 --> 00:29:15,039
this to you guys any further than these are in fact 24 gab memory modules and am

392
00:29:13,000 --> 00:29:19,200
I am I allowed to run them like in my system is there any reason I couldn't

393
00:29:17,360 --> 00:29:21,519
yeah just can't don't sell them and these are

394
00:29:21,919 --> 00:29:28,440
XMP bleep that oh bleep that

395
00:29:25,279 --> 00:29:30,039
too all right then I hope you guys

396
00:29:28,440 --> 00:29:33,679
enjoyed this behind the scenes look I want to give a massive shout out to

397
00:29:31,640 --> 00:29:37,840
Micron for making this possible this is not an easy thing to pull together and

398
00:29:35,919 --> 00:29:43,320
we had a ton of fun but I don't want to let that distract from the serious work

399
00:29:40,039 --> 00:29:44,600
that they're doing here this wall is all

400
00:29:43,320 --> 00:29:49,440
of the 50,000 patents that Micron holds in

401
00:29:47,360 --> 00:29:53,240
storage and memory and Associated Technologies this is literally the

402
00:29:51,399 --> 00:29:57,440
largest R&D facility for memory and storage products in the Western

403
00:29:54,399 --> 00:29:59,640
Hemisphere and it's pretty clear to me

404
00:29:57,440 --> 00:30:04,080
based on seeing it that no matter what your budget is Idaho has everything you

405
00:30:02,120 --> 00:30:09,399
need for your next PC be it a top-of-the-line SSD and memory kit from

406
00:30:06,320 --> 00:30:10,880
crucial or a literal potato these

407
00:30:09,399 --> 00:30:14,679
stickers are great but literally everything about them is a Lie from the

408
00:30:12,760 --> 00:30:19,360
part number to the fact that it says crucial on it that's their consumer

409
00:30:16,840 --> 00:30:24,000
brand and no consumer is ever going to touch these modules I have been

410
00:30:21,240 --> 00:30:28,679
specifically instructed to destroy these if I ever take them out of my computer

411
00:30:26,159 --> 00:30:33,000
because they were not supposed to leave this lab if you guys enjoyed this video

412
00:30:30,960 --> 00:30:36,320
you might also enjoy our Intel Fab tour where we went into a little bit more

413
00:30:34,480 --> 00:30:41,440
detail about the actual lithography process that one's also wicked cool