What is Chip Architecture - And Why Is It the Stage Where You Decide What the Chip Will Really Be?

In previous posts, we talked about what a chip is, what an SoC is, and how you “write” hardware using RTL. Now we’ll approach a central part of the Frontend world:

Chip Architecture

This is the stage where system designers determine:

• Which units will be in the chip
• How they’ll communicate with each other
• What resources are needed
• What the data flow looks like
• And what the chip’s overall capability will be

Simply put: Architecture is the top-level design - the plan that dictates everything that comes next.

Why Do We Even Need Architecture?

A chip is not a “single component”. It’s a complex system that contains:

• Computing units (CPU / GPU / NPU)
• Memories
• Controllers
• Internal communication bus
• Support units (video, network, etc.)
• I/O protocols
• Clock and power management system

All of these must work together in an organized manner.

Without architectural planning:

• The system won’t be efficient
• There will be resource conflicts
• Performance won’t meet requirements
• It won’t be possible to verify or build it

What Does a Hardware Architect Actually Do?

1. Understands Product Requirements

What does the chip need to do? Is it intended for a car? Camera? AI accelerator? Mobile? Server?

Each product - different requirements.

2. Determines Which Blocks Are Needed

For example, in an AI chip:

• Matrix accelerator
• Fast internal memory
• Communication controller
• CPU for management
• Compression/decompression engine
• I/O for running the model

3. Defines Communication Structure Between Units

Who talks to whom? What’s the bandwidth? How are loads scheduled?

4. Determines Memory Division

How much Cache? What internal SRAM? How does each unit get access?

5. Defines Clocks and Power Consumption

Who always works? Who turns off in power-saving mode? What are the required speeds?

Data Flow - One of the Most Important Things

Architects diagram:

• Where does data come from?
• Who processes it?
• Where does it go next?
• Where are bottlenecks created?

This flow is critical for performance.

For example:

If an NPU produces 200GB/s of data but the bus can only transfer 100GB/s - The chip won’t meet requirements.

The architect is the one who prevents these situations in advance.

What Comes Out at the End of the Architecture Stage?

A long and detailed document that serves all development teams:

• Block Diagrams
• Communication Interfaces
• Clock schedules
• Memory requirements
• Logical description of operations
• Division of responsibilities between teams

This is the foundation on which RTL engineers will start writing the hardware.

Why Does Architecture Come Before RTL?

Because without planning:

• You don’t know which modules to build
• You don’t know how they should communicate
• You don’t know what the requirements are
• You don’t know what’s important and what’s secondary

Architecture is the “story”. RTL is the “chapter”.

An Illustrative Analogy

Imagine you’re building a house:

The architect decides:

• How many rooms there will be
• Where the kitchen is
• Where the bathrooms are
• How air will flow
• What the power consumption is

The RTL engineer takes the plan and builds the details:

• How exactly each room looks
• How to build each wall
• What will be in each corner

Without architecture - there’s no plan. Without a plan - you can’t build.

Summary

Chip architecture is:

• The top-level definition of what the chip does
• How the units within it are arranged
• How information flows between them
• What resources are needed
• And how everything works together in an actual system

It’s the stage where you draw the big picture - before entering the logic lines in RTL, before simulations, and before physical construction.

---

This is the “Chip Design Journey” series - an in-depth journey into the world of chip design. In the continuation, we’ll learn about Verification, Synthesis, Place & Route, and more critical stages on the way to an actual chip.