Bay Area, Micro-Nova Semiconductor Phase II Project.
Looking at this bottomless pit that had devoured Gamenova's recent profits in one fell swoop, even Lin Lixin couldn't help but feel a pang of heartache.
Let's not talk about the equipment.
The upgrade of the cleanroom alone cost over 450 million, even approaching 500 million US dollars.
Five hundred million!!!
This is truly throwing money into a fire.
Nexus burned through its quota for more than half a year, allowing wafer foundries to catch up completely in one go.
There has long been a misunderstanding about the semiconductor industry.
It seems that as long as you have an advanced lithography machine, you can create advanced chips out of thin air.
But that's not actually the case.
In a sense, lithography machines, which are readily available equipment purchased directly from manufacturers, are actually the easiest part to solve.
Upgrading cleanrooms is an unavoidable part of the process. This upgrade aims to advance the entire production line to 0.8-micron process technology, and some high-precision production lines will even attempt to challenge deep submicron process technology. The first phase will be responsible for producing inexpensive chips using 1-micron and 2-micron processes.
What does 0.8 micrometers mean?
To put it simply, the width of a single human hair is approximately 70 micrometers.
If even a single hair appears on the production line during processing, the entire wafer can be scrapped.
This kind of thing is alright.
At worst, we can find a bunch of bald guys or people with hairnets.
But you can protect yourself from hair, not dust.
In a standard, normal office environment, a single dust particle in the air can be nearly 20 micrometers in size.
Falling onto a wafer can also have catastrophic consequences.
If even a single speck of dust falls onto a standard 6-inch wafer, it could cause dozens of chips to fail simultaneously.
If you don't upgrade your cleanroom, you'll be waiting for your product yield to plummet.
If yield rates don't improve, chip prices will skyrocket, and Micron as a whole will go bankrupt. Not to mention, the yield problems with blue LEDs and LCD screens haven't even been resolved yet.
After deducting the expenses for the cleanroom, the remaining costs for equipment procurement, manpower, process research and development, software, and other expenses totaled nearly two hundred million US dollars.
The cash flow that had just been injected into Pokepal and Google's business was completely drained in one go.
The investment from their previous games was also poured in.
"It's truly a bottomless pit."
Just calculating the subsequent expenses gave him a toothache.
Keep in mind that this isn't even a modern nanoscale wafer fab.
This is just a deep submicron level wafer fab.
One has to marvel at the rapid pace of scientific and technological progress.
Who could have imagined that advancements in semiconductor technology would be so concentrated in just a few short years after the 1990s?
"Well, well, it's a path we have to take anyway."
No matter how expensive, we have to invest. Semiconductor technology can only be developed step by step.
On the bright side, at least every penny spent was worth the ticket price.
The 1990s were a period of rapid growth in demand for semiconductors, coinciding with the dot-com bubble burst.
By simply doing a few OEM/ODM projects and selling some hardware, you can make your money back sooner or later.
Compared to the true industry leaders like Intel, IBM, and NEC, our own expenditures are not particularly high.
"Don't be stingy. Phase II involves the entire i-line photoresist line. Ideally, even if we break through to the deep submicron level in the future, we will only need to upgrade the cleanroom."
Vivian walked up to him and teased him.
Prior to this, Micro-Nova Semiconductor used g-line, which is a light source with a wavelength of 436 nanometers, almost exclusively across its product line.
It works incredibly well in manufacturing environments with processes exceeding 0.8 micrometers.
It is low-cost, technologically mature, and stable.
However, below 0.8 micrometers, or even at the 0.5 micrometer level, this thing becomes less reliable.
The first-phase factory wasn't incapable of producing chips using a 0.5-micron process. However, the yield was shockingly low, rendering it completely devoid of commercial value.
At this stage, G-line is completely unreliable.
Therefore, i-line with a shorter wavelength becomes the most economical choice.
The so-called g-line and i-line actually refer to one of the spectral lines of a high-pressure mercury lamp light source.
A light source of a specific wavelength is filtered out using a filter.
The image is focused onto the photoresist using a specially coated high-precision lens.
It is not difficult to see that a simple modification to the light source wavelength has led to the need for a complete redesign of the entire lithography system.
From the lens to the coating to the photoresist to the mask, and even the software system, nothing can be reused.
This is another major expense for wafer fab upgrades.
"It's expensive, but the benefits are obvious."
Lin Lixin was quite open-minded; he said he felt sorry for the situation, but he also understood the necessity of the matter.
Upgrading to the i-line level means that for a long time to come, wafer fabs will not need to consider replacing equipment. i-line will remain competitive even up to the 0.35-micron deep submicron process.
Even after the turn of the millennium, mid-to-low-end chips and LCD panels using i-line technology remained the mainstay of the consumer market.
Even if he wanted to create the next level of deep ultraviolet (DUV) and extreme ultraviolet (EUV) radiation, no one on Earth could do it now.
If it really needs to be done, Sage Lin will have to start from scratch and do it entirely by hand.
With a complete upgrade to i-line, they can now directly migrate their existing chips.
Even the design doesn't need to be changed; just scale the layout, and it's ready for production.
Previously, a single wafer could only produce 100 chips, but with the new process, the same wafer can now be cut into more than 150 chips.
Not to mention the significant increase in the yield rate.
In terms of cost-effectiveness alone, it represents a truly leap forward.
"The current plan is to relocate Athena and Novaradeon's operations here, oh right, and our memory chips as well."
Vivian pondered, her heart also filled with excitement.
If we're talking about who benefits the most from the process upgrade, it's undoubtedly memory chips. Storage is all about density.
The higher the density, the greater the capacity; this is self-evident.
With current technology, theoretically they could easily produce a large number of megabyte-level memory chips, even 16MB.
That is, the 2MB level.
"Memory will definitely need to be migrated, but it would be a waste to migrate the processor and graphics card. I will design a second-generation architecture for them."
Lin Lixin shook his head.
New technologies bring new possibilities.
Especially now, he is different from when he designed Athena.
The evolved [ Power of WAAAGH! ] can help him easily reverse engineer various semiconductor architectures.
Adhering to the idea that a small investment doesn't count as a big investment, Lin Lixin conducted some simple investigations and studies.
"I recently disassembled several flagship processors from MIPS and SPARC, oh yeah, and Motorola's too."
In this day and age, those who can work for a few leading chip manufacturers are all among the best of humanity. Their ideas are the most valuable Fortune.
By disassembling these processors, a great deal of valuable knowledge can be reverse-engineered.
"I think Intel's approach is somewhat instructive, and we should also add L1 caching to Athena."
The new manufacturing process results in SRAMs that are smaller than ever before.
Small enough to be integrated into a processor.
It's small enough to fit a full 16kb of L1 cache into the processor.
16kb!
It's worth noting that the Nintendo Famicom, which was a global phenomenon a few years ago, didn't have this much RAM and VRAM combined.
The Famicom (FC) console only had a paltry 2KB of memory.
But even at that time, this was considered a high-end configuration.
Now, the processor's own L1 cache alone can reach a full 16kb!
Memory has long been at the megabyte level. "Add caching?"
"Yes, and it has a very large cache. Vivi and Athena are processors specifically designed for game consoles."
Lin Lixin nodded.
This has always been a shortcoming of the athena, and it was a helpless measure due to the limitations of the technology in the past.
"For video games, the biggest bottleneck isn't actually computing power, well, at least that's true for most games."
Many games don't actually require much computational effort.
I spend most of my time doing two things.
Graphics chips and memory.
There's nothing that can be done about the graphics; the only solution is to upgrade the graphics card specifications.
As for the second point, there are solutions.
The simplest solution is to upgrade the memory specifications; make it larger, faster, and with higher frequencies, which will naturally make it more flexible.
But now, another path lies before Lin Lixin: L1 caching.
The CPU can have its own cache, and even a large number of very large registers.
This means that the CPU can directly store data in its own cache, without needing to frequently exchange data with memory.
A round trip like this can save a lot of time.
For memory-intensive video games, this improvement is readily apparent.
Only data that is swapped less frequently and used relatively infrequently will be loaded into memory.
More static resources are left directly on the hard drive.
"But wouldn't that require a lot of instructions for data exchange?"
The processor is a very busy person, constantly processing a large amount of data.
As data volume and storage space increase, the burden on the CPU for this part of the business is also becoming heavier.
In particular, Athena itself is a native heterogeneous computing architecture.
In Athena's working environment, each domain, such as memory, graphics card, optical drive, and hard drive, has its own cache space. Even peripherals like steering wheels have their own small cache.
"You're right, so I took apart the MIPS R4000 to do some research."
Lin Lixin was amused.
The MIPS-R4000 is the same chip used by 3DO that is touted as the world's first 64-bit processor.
"We can add an 8-channel DMA to the processor and let it handle all the data exchange, while the processor only issues commands."
This feature is absolutely designed to target video games.
Frequent and extensive data exchange is a problem that is difficult to avoid in video games.
With DMA handling this part, the processor can free up a lot of resources to do more important things.
In addition, branch prediction and multiply-accumulate circuitry, which were previously impossible due to process limitations...
athena-ii remains unchanged and cannot be used for general computing.
But in the world of video games, she would be an absolute king, years ahead of her time.