738880_en-US

liliconcom

CMP Two-Polish Table Optimization for Max Tool Output - (2023-US-PO-1446)

Chemical mechanical planarization (CMP) is a process used in the semiconductor industry in which a layer of the chip is planarized or flattened in preparation for putting another layer on top of it. The chemical part consists of chemical reactions occuring between the liquid polishing slurry and the top layer of the substance being polished, which is often softened in the process. The mechanical part occurs after the chemical softening, removing the top layer and exposing the next atomic layer for chemical reaction. The polishing is crucial to ensure that the subsequent layer doesn't have electrical opens or shorts due to its topography on the preceding layer. Many CMP processes consist of bulk polish on one table and then a final polish on a second table. This two-table process helps achieve the maximum possible output from a CMP tool if the two tables are properly balanced for bulk and final polish times.

I 'm still seeing your Polish Table 2 Throughput Optimization slide too ,

by the way .

I 'm not sure if it 's just really laggy or what 's going on there .

You 're not seeing the title slide ?

Now I 'm seeing the title slide .

I 'm going to go ahead and unmute myself .

Yeah , I don 't know if there 's anything we do about that either .

That 's got to be like a Zoom or potentially an Internet thing .

Well , let 's go ahead and do this

and then if I see that things look really janky ,

I might be messaging you .

But I think for the most part ,

if you just maybe take a quick pause between slides ,

we 'll be okay .

I 'll go ahead and mute ,

and then you can go ahead and get started , Tom .

I 'm really glad to be here today and presenting to you .

I 'll be discussing Chemical Mechanical Planarization or CMP .

This is an integral step in the semiconductor manufacturing process

that I will give a little bit of background on

as I move through my slides .

Let 's go to the first slide of the presentation .

Quick overview of CMP .

Once again , that stands for Chemical Mechanical Planarization ,

and that describes the two processes involved

with planarizing certain layers of circuits in the semiconductor field .

It 's crucial to planarize each layer of circuits

so that the one that goes on top of the preceding layer

will not have any shorts or opens or short circuits .

The chemical part of chemical mechanical planarization

is when we introduce a liquid

to react with the top atomic layer of the material that we are planarizing .

That reaction softens the layer ,

whether it 's a metal or an oxide , whatever the case may be ,

we choose that chemical so that it will soften the top atomic layer .

Then the mechanical part comes

with the abrasives that are carefully selected for that polishing slurry .

These abrasives then remove that softened top layer ,

exposing the next atomic layer to be softened by the chemical reaction .

This process is repeated atomic layer after atomic layer

until the desired level of planarization is reached .

Like I was saying ,

this planarization is crucial to do in between each layer of circuitry

to ensure that we have solid electrical isolation

without any shorts or opens .

This process enables us to pile many layers on top of one another

and reduce the footprint of whatever chip we 're making .

When we do the CMP processes ,

many of them consist of a two polish table process to enable maximum throughput .

Typically , the first table consists of a bulk polish

where most of a substance is planarized and removed ,

and then a final polish or second polish table

where we do the final touch -ups

and ensure that we have good electrical isolation between the circuits .

Each polish table has a polish pad

where we press the wafer containing the chips against ,

and this enables the polish process to occur

and it enables the slurry to be pressed against the wafer that contains the chips .

The polish pad , unfortunately ,

it wears down as each wafer is polished on it ,

which inhibits the transport of this chemical slurry to the wafer surface .

Typically , as the polish pad wears , we see a reduction in polish rate ,

which causes longer and longer polish times .

If we don 't take that wear factor into account

and we 're setting up the two -table process ,

then we will have one polish table begin to predominate ,

and it can actually develop such a long polish time

that it will become a limiting factor in our throughput .

For my presentation today ,

I 'm explaining a process whereby we can take into account the pad wear rate

and still ensure that we have optimized throughput .

Approach and the results .

I gathered and analyzed data in JMP

to show the polish time versus polish pad life relationship

for tables 1 and 2 for a CMP two -table process

at one of Texas Instruments Fabrication Facilities .

I then used JMP scripting language ,

the built -in integration function to calculate the area

under the polish time versus pad life curve for both tables .

Then , finally , I used the system of equations

to solve for a Table 1 polish time starting point

that enabled equal polish time versus polish pad life integrated areas

for tables 1 and 2 .

This ensured that we had achieved a maximized tool output

for the CMP process .

Conclusions are that we can use JMP

to develop , first of all , polish time versus polish pad life curves .

Then , second , we can use JMP scripting languages , built -in integrator .

We can use this combination

as an efficient way to optimize CMP two table processes for throughput .

Let 's go to the next slide .

This curve that we 're looking at is what we get when we plot the polish time

on the first table versus the polish pad life of that first table .

As we can see , the polish time as the pad wears

and we polish more and more wafers on this polish pad ,

we start to see polish times increase quite dramatically .

Using JMP , we can fit a line to this data

and develop a simple Y equals Mx plus b slope intercept function to fit this data .

I 'm using a first order here ,

but if the data justified it , we could use a second or third order .

But in this case , a first order is sufficient .

We developed the equation for polish time versus pad count ,

and then we use the JSL integrator to integrate the area under this line .

That 's for the Polish Table 1 .

Then I 'll talk a little bit more about the equations needed to solve

for the optimal settings .

But first of all , we 'll jump to Polish Table 2 .

Polish Table 2 is the final , it 's the finishing table

where we finish removing whatever the bulk material is

we 're using to isolate or form these circuits , electrical circuits .

The same thing for Polish Table 2 ,

we developed the curve from historical data

of polish time in seconds versus the polish pad life .

Typically , for the polishing plat , for the final plat or final table ,

the trend with polish pad life is not as pronounced as it is

for the bulk for Table 1 .

Before discussing a little bit more about these equations ,

just to make sure everyone 's on the same page ,

I wanted to show what physically this looks like on the wafer surface

for doing this two -table polish process .

Before we start to polish anything ,

we have a wafer with pattern circuits on it .

Then , on top of that wafer ,

we 've deposited a large amount , relatively speaking ,

of whether it 's isolator or electrically conductive material ,

whatever is needed to form the circuits that we 've etched here .

After Polish Table 1 , also known as the bulk polish table ,

we 've removed a significant amount of this bulk material

and we 've planarized it , made it very flat .

Then after Polish Table 2 ,

which is also called the clear or the finish table ,

we 've removed all of this bulk material

except where it 's actually needed in the pattern circuits .

Physically , that 's what 's happening in this two polish table process .

Let 's go back quickly to the equations .

We integrate the area under Table 1 , which is on the preceding slide .

That 's the bulk polish time versus Table 1 polish pad count .

Then we also integrate the area under this curve ,

which is the final or finish table .

Same thing , integrate the area under the polish time versus pad count line .

Then we solve the linear system of equations to find which Y -intercept ,

in this case , it 's b1 , and that 's for the Table 1 .

We solve for that b1 that will create an equal integrated area

under the Table 1 and the Table 2 curves .

Just as a reminder ,

the JMP scripting language has a handy function

for enabling quick integration

and determining the areas under these two curves .

Then based on a few other properties

that we know based on historical or baseline analysis ,

then we 're able to set up a system of equations

with four unknowns , four equations ,

and then we solve for that intercept on the first curve

that enables maximum throughput in equal area under these curves .

With that ,

almost the entire process can be done in JMP ,

with the exception of solving a system of equations

that can indirectly be done in JMP , or you can write a script .

But I 've found it easier just to solve the system of equations manually ,

but it could easily be set up to be done in JMP as well .

When you finish this process ,

then you 're able to optimize this two -table CMP system

and you can maximize the throughput

and so minimize the number of polish tools required ,

which saves a lot of money and saves a lot of time .

That is the end of my presentation .

Thank you .

Advanced Statistical ModelingBasic Data Analysis and ModelingData AccessData Exploration and VisualizationQuality and Process Engineering