
zzattack
Members-
Posts
43 -
Joined
-
Last visited
-
Days Won
17
zzattack last won the day on November 25 2024
zzattack had the most liked content!
Recent Profile Visitors
The recent visitors block is disabled and is not being shown to other users.
zzattack's Achievements
-
I have hundreds including prepopulated ones, you can send a private message on here if you like. I ship from The Netherlands.
-
I do wonder where he managed to obtain the Marvell ic in reasonable quantity.
- 115 replies
-
- xeniummods
- v3.x
- (and 4 more)
-
That's correct. The interposer works very well, every upgrade I've performed with them so far has worked out perfectly. If anyone wants some boards, I have 250+ and am sure I'll never use them all myself.
-
Another clone. Credit to original sickmods design, this is electrically 1:1 with the addition of pins for the programming header on the right side. Contrary to all the other clones this one actually publishes kicad schematic + board layout, instead of a PNG of the .sch and gerber output, since I really do not care if someone wants to modify this or put their name on it. https://github.com/zzattack/xerc2/releases/tag/v1.0
-
Yes, on the xbox motherboard. lOgIcAl was asking about the interposer pcb.
-
Normally yes. This is why in the rev2 design no via in pad is used and I've made a considerate effort to keep vias away from silk screen. However, I then found out that jlcpcb charges no extra for POFV: : https://jlcpcb.com/blog/32-Free-Via-in-Pad-on-6-20-Layer-PCBs-with-POFV In this case, this is very attractive so I made rev3 specifically to take advantage of this.
-
Small necro-bump on this project. About a year ago I had started to 'delayer' a n64freak interposer for the purpose of revealing its inner layers, allowing me to scan them for reference and potential recreation. Being unfamiliar with KiCAD there was some interest in learning about it, so as a way to build up experience I began to redraw the schematics from the original FriendTECH. This did not take excruciatingly long since it is mostly a point-to-point hookup between the two CPU footprints, with only very minor circuitry added. I completed the schematic, but after speaking with Kekule who had done a similar exercise I did not persue this effort further, since his files would eventually be released. Then, when the Kekule interposer gerber release was announced 2 weeks ago, I felt somewhat disappointed that 1) I never completed what I started and still hadn't used KiCAD for layout, and 2) neither schematic nor pcb layout files were released along with the manufacturing files. Therefore I set out to take a stab at resuming where I left off. Initially I overlayed the delayered scans with the positioned footprints and began to trace them into the copper layers. This went well for the 45° angled signals but eventually grew old quick. Later I found a way to convert the Kekule gerber files into editable paths which I could place into the layout, and after annotating them with the correct schematic nets I ended up with a completed board reasonably quickly. This was to be revision 1, yet another clone of the 20 year old design, now with the benefit of having the actual editable design files along with them. Oh, and the useless clock reduction circuit is chopped off. Now the original design has the major drawback that the BGA footprint is in the corner of the board, which causes one of the original heatsink bracket mounting holes to be covered, and during reflow the uneven distribution of the weight can cause the interposer to tilt/slant during reflow. In the MakeMHZ discord, user 'doom' had flaunted some pictures of his interposer which was not a clone but a full remake. It looked very very clean, did not have the BGA footprint in the corner, and seemed like an overall massive improvement over the original. My next challenge would be to offset the BGA pads in a similar way. This would of course require that I re-route the entire board. I was told by Kekule that the FriendTECH routing was absolutely next level and despite him having a lot of PCB engineering experience, he could not get the trace lengths to match up within the requirement of 1/10th inch for all of the data lines. This withheld me from taking a stab at it myself. I've done some PCB design up to 4 layers, but never considered myself expert-level, so I expected this to fail quite miserably. After putting my kids to bed at 9pm I sat down and didn't get up until it was fully routed at 3am. In about 6 hours I drew up what I thought would take me weeks, if I could even do it at all in the first place. The result of that was uploaded as 'revision 2'. I later cleaned up some of the silk screen, ensured no silk screen is printed over the vias, optimized the power planes a little bit and placed an order at JLCPCB for revision 2.1. The layout I drew up conforms to quite liberal design specs with 5 mil traces and 0.3mm drills and does not make use of via-in-pad technology meaning that to breakout the BGA signals, a small bit of trace is required and via's are placed in the center of 4 surrounding pads. This complicates routing of course since more real-estate is required for every pad. The advantage of using such design specs is that board houses typically start charging extra for smaller drills, so it can be produced very cheaply: there's a $2 for 5 offer on JLCPCB. The revision 2.1 board looks very messy compared to the aforementioned clean board made by doom. Looking at it closely, he seems to have made use of via-in-pad technology, allowing to directly break out from the SMD/BGA pads into the desired signal layer. Figuring this would take a ton of money to have produced I looked up a quote at JLCPCB and found that the additional premium is very reasonable actually. I'm considering doing another re-route where I'll aim for a really clean layout while making use of this additional capability. For now, here's the true open source designs: https://github.com/zzattack/xbox-cpu-interposer Thanks for reading
-
So cerbios formatted scheme remains unsupported?
-
I recently got asked for some tips on this upgrade and figured others could benefit from the info as well. So for posterity, here's what I recall as being the most relevant info, condensed. There's 2 important topics to get right which require some strategy: alignment needs to be dead-on, and the interposer needs some help to be mounted flat. I'll explain why. For alignment, generally you can be just under half a pitch off with your BGA alignment and it'll work out fine when the chip settles. So, the BGA pitch on this motherboard of 1.25mm suggests alignment is very forgiving. However, vias centered within groups of pads are not nicely tented, and instead take up most of the leeway you'd normally have with alignment. I'd say you don't have ~0.6mm of tolerance, but rather about 0.15mm. To assure I get good alignment, the first thing I do is eliminate the possibility of bridges forming between the BGA pads and nearby vias. You can simply tent these vias manually using UV curable solder mask. Takes about 10-15 minutes, and with this, a lot of alignment margin is recovered. At the same time, I place some 0603 resistors near corners of the BGA array and interposer outline. See the red marked circles below. The BGA pads on the underside of the interposer do not cover the entire interposer area, so the interposer tends to tilt/slant during reflow because it's gravity center is off. When the interposer is tilted, BGA balls can get too compressed and are unnecessarily stressed. Now the 0603 resistors just happen to have a height very similar to properly compressed 0.76mm solder balls, so this very simple approach ensures the interposer will get mounted flat. Next, my definitive way to get perfect alignments, after a number of less precise prior attempts: I drilled through the corner BGA pads on the underside of a sacrificial interposer. See circles in picture below. The resulting holes allows to see the xbox BGA pads underneath, and thereby align the sacrificial interposer perfectly. Then, I tape it down temporarily. The next step is to fix this interposer in place. The idea is that by placing e.g. 0805 capacitors along the edges of this interposer, we can lift it up and replace it with a fresh interposer in exactly the same position. Furthermore, because the interposer cannot move laterally, the interposer won't "dance", and therefore balls cannot join with the vias centered within groups of pads either. With the alignment also being spot on, there shouldn't be a "snap" when the BGA aligns itself due to surface tension either. I believe this thoroughly eliminates everything that could potentially yield faulty mounts. The very last optimization I added was to place a single case screw diagonally over the interposer as depicted above. This counteracts the weight imbalance due to the BGA pads being off-center underneath the interposer. The positioning I found is determined experimentally and results in the interposer settling down perfectly straight, i.e. all 4 corners go simultaneously. This isn't super important since we already guarantee flat mounts due to the 0603 resistors acting as standoffs from a prior step. Running short on time for this writeup so I'll conclude with the only other vital step: use a bismuth compound solder paste for the CPU itself. Bismuth solder has much lower melting point than normal leaded, so this allows soldering of the CPU without inadvertently reflowing the interposer (but with a lot of added weight on top, potentially squashing the balls!).
-
Well, I finished my batch and moved on. Maybe stellar will revigorate some interest. The only one I've got left is spoken for but I might do another few when I can get my hands on some stellars.
-
Bowlsnapper started following zzattack
-
I've found my interests gradually shifting away from the xbox cpu convresion. That's not necessarily all bad, it mostly means I've achieved what I wanted. Both cooling solutions (glued/screwed) perform fine and I believe all of the kinks in my process have been ironed out. I did build up a nice stack of consoles to convert, and have indeed converted a fair bunch of them. A few sold through ebay quite easily, and a few through discord/twitter. I still have a few more available, maybe there's some interest here, so I'm putting two up for offer here. I'd like €250 + shipping for the one on the left, 1.4 revision with glued heatsinks. The one on the right is a 1.0 with screw mount heatsinks and open xenium priced at €300. Price excludes shipping. Both are complete systems with dvd player, hdd and shell. The 1.4 is 230V and the 1.0 can be sold as either 110V or 230V on request. Aside from the CPU upgrade they have 128MB of RAM, are recapped and the board is ultrasonically cleaned.
-
Haven't got around to setting that up, but the first few have sold. They run off 230V but swapping the PSU is al that's needed for making them work on 110V. I'm still actively looking for ways to improve the heatsink mount. They are fairly difficult to position entitrely flat on the cpu heatspreader.
-
SoftMachine started following zzattack
-
Not exact numbers, but close. Didn't keep a record since in the end, they are all fixable. Unopened, working OOB: about 10 Opened before, but still working: about 7 Unopened, slight defects (front panel buttons/clock cap damage/poor DVD): 3 Opened, slight defects: 4 Unopened, not working at all: just 1 Opened, not working at all: just 2 The 'not working at all' were actually easily fixed by capacitor replacements. In fact, the "slight defects" are more work to fix because it usually involves trace repair due to supercap leakage. Two DVD drives ended up having no laser light at all. Some others required slight pot meter tuning.
-
They're traded at very low volume so estimating their value is hard. But the point about rarity has some merit. When I'm dumping 20 at a time, it's clear I'm not expecting all, or even any of them to fetch top dollar. That said, I've carefully deliberated the price point at which I want to offer them fully kitted with recap + clock reduction circuitry + 128MB upgrade and cleaned shell + dvd drive + hdd. I'm not exactly sure how much deduction is feasible when it's just the bare "motherboard with upgraded CPU" and none of the rest. But it does save a lot of time. And it allows me to discard non-working parts instead of fixing them. On the other hand, if everyone ends up buying just the board because it's cheapest and they'll either not care for or do the other upgrades themselves, then I'll have a ton of cases to discard of, which is also not ideal. I'll decide on that soon, once I get some e-commerce thingie going on a website. I think it'll be around €275 for the bare upgraded board, and it'll save a fair bit on shipping too probably.
Board Life Status
Board startup date: April 23, 2017 12:45:48