"Hello, teachers, everyone, thank you very much for joining our EDA project team with trust in me. Today's meeting will briefly explain the design concept of our minimalist EDA, and we have already divided our work on everyone's next work."

In the small conference room on the fourth floor of the laboratory, Ning Wei was giving a report to everyone in the project team.

There is still pressure.

After all, in addition to Yu Xingwei, the youngest ones are also older than him.

"So, the EDA software that our project team will design is not as complicated as everyone imagines. I call it object-oriented EDA software design."

"I'll assume here that each structure that can be made by the existing technology is an object. I've divided these objects into four types."

"They are respectively meta, that is, the smallest structural unit that can be made by the existing technology, which can be updated with technological breakthroughs."

"A group, several elements are connected to each other and arranged to form a structure with a single function.

"Groups, several groups with the same or different functions are connected to each other, and are arranged to form a structure that can independently complete a certain function."

"Container: a structure used to support units, groups or groups."

"Each object must contain two parts: attributes and functions."

"Attributes not only contain the physical properties of the object, such as size, size, etc., but also the index and representation properties of the object, such as representative symbols, unified sequence numbers, etc., and also include the layout properties of the object, such as connection methods and arrangement methods. The function must have input parameters and return values, which are descriptions of the object's functions, such as single-item conduction, amplification, cutoff, etc.."

Ning Wei compared the PPT that was kicked out yesterday and expressed his basic design concept to the project team members.

"From this perspective, everyone should see that our EDA design idea is actually very simple. On the one hand, we need to make the operator easily position, arrange and wire symbols representing different types of objects on the design interface; on the other hand, all operations of the operator are actually called in series by the computer background of the computer to form a large function process."

"The above are requirements for the design process. At the same time, our EDA software must have simulation functions. The EDA simulation process is a process of inputting parameters into the large function process and recording feedback results, and also includes monitoring of the parameters and return values ​​passed between each subprocess.

"The EDA instantiation process requires the layout position and arrangement of objects to generate running scripts to control the lithography machine for production and manufacturing."

"Based on the above requirements, we directly guide the software structure."

"Object library: Used to define existing objects and save objects for new designs in the future."

"Function library: Functions used to save existing or future new objects."

"Design interface: used to layout, arrange and connect objects."

"Function Connection: Functions used to concatenate and call functions of arranged and connected objects in a row and connected manner."

"Simulation and monitoring: used to send parameters to structures and detect the return values ​​of each function node."

"Production function: used to script the layout, arrangement and connection of objects to the lithography machine for production."

"In addition, because this is a brand new EDA software, we cannot rely on peripheral interfaces to introduce plug-in programs, so we also need a library that is different from other EDAs. This library corresponds to the object library. Every time we design a structure, we need to update an instruction set and supporting functions for this structure."

"In other words, every chip we design for a specific problem must have all-round support. This is also the place where our product differentiated competition lies. Our products are not only aimed at EDA designers in specific large factories. My original intention is to enable a primary school child to use our software to design useful chips after reading the manual."

"After our software is launched, the structure and supporting instruction set designed by professionals, and the functions can be synchronized into our library after verification and review."

"This also puts forward a requirement for our products, which is to build a logical judgment ability on the structure. Simply put, what you want is what you get. In other words, users only need to be able to describe the requirements more accurately, and our software can automatically recommend the architecture and supporting configuration files that can complete this requirement!"

"I think everyone has already understood the design concept of this software. It is simple, and it is simple enough, so it almost does not require any promotional costs."

"Do you have any questions?"

...

The members of the project team looked at each other in surprise.

Because it is hard for people to imagine what the significance of developing such an EDA software is.

Do you use it as a toy for children?

That’s right, simplicity can save a lot of promotion costs, but it also means that this thing is not practical and valuable!

First of all, it is almost impossible to design and simulate large and complex general-purpose chips, such as CPUs and GPUs.

The other is the application layer.

Ning Wei means to bypass the existing various architectures and instruction sets, and also involve compatibility issues.

In fact, at this stage, AI chips already have ready-made solutions, such as FPGAs.

Unlike Ning Wei's idea, FPGA can be directly understood as a universal chip with a unified structure.

Users use the hardware description language to design the hardware circuits of the FPGA by burning the FPGA configuration file. Every time a burn is completed, the hardware circuits inside the FPGA have a certain connection method and have certain functions. The input data only needs to pass through each gate circuit in turn to obtain the output result.

This method is obviously more cost-effective than the ASIC model designed by Ningwei. Although FPGAs also have their own shortcomings, such as their performance is not as good as ASIC chips, the price is also higher than ASIC chips.

The reason is that various artificial intelligence deep learning algorithms are in a period of rapid development and iteration is extremely fast. If the corresponding optimal configuration files are designed according to Ning Wei's idea, it means that every algorithm is changed, a new generation of dedicated chip structure must be updated in the object library.

Once the update is slow, it will be out of date...

So few people in the entire project team are optimistic about the prospects of this EDA software.

But everyone's ideas are different.

Some people think it doesn't matter. Anyway, after the funding is allocated, everyone can make some money no matter what. As for whether this software has practical value, there is nothing to care about.

Anyway, it’s a special grant, so it’s not useless.

But some people think this is inappropriate.

The money from the public is also money. Now the lab plate is so big. If they use too much on their projects, other projects will have to reduce funds.

What should I do if I spend this money on projects that are destined to be of little value?

So after silence, someone made a different voice.
Chapter completed!