Intel shows Meteor Lake test chips with stacked design – Computer – News

You want to be able to put as many chips on a wafer as possible because of the costs. So by baking the chiplets on top of the chips you need less ‘footprint’ I think.

no, that is not true, ie: yes, if that were the only criterion then you were indeed right, because more cpu per gram of material would then be cheaper per end product.

however, this is not the case in practice. other factors generally influence this type of production line much more than just the material used.

1. the more complex the design, the more error-prone the process: so the more broken chips end up on the waste mountain.
2. the more complex the design, the more likely your chip machine won’t be able to finish it in one round, and so how much extra time (on the machine) you need per chip, more time is less chips = more expensive, more time = more power = more expensive. more laps on the machine more chance of errors when realigning on the machine = more production errors. (And so on)

why then,
From the little bit that I understand it goes something like this,

each path from part 1 (for example cache) to part 2 (for example computing core) is made of a conductor, which conductor has a certain resistance per linear nm. the longer the path that such a signal has to travel, the more resistance has to be bridged AND THEREFORE the stronger that current must be.

the stronger that current is, the more interference it will give to other parts of the chip and other currents in that chip… the more interference there is, the further apart components have to be in order not to suffer from that interference. .. but also, the longer those lanes become again and so the currents have to become stronger again.

there are really only 3 ways to solve this endless race for more and more power and increasingly complex ways to counteract that interference;

1: lowering the resistance on those lanes (by, for example, superconductivity), = not yet possible at the moment

2: making use of the properties of machinism and how they can cancel each other … that will make your chip endlessly complicated.

or 3. and intel seems to be choosing that now, you come up with a way to shorten those laps. and if not anymore. left, right, front or back, then go up or down.

[Reactie gewijzigd door i-chat op 19 november 2021 12:09]

–