Let's compare real-world examples of Von Neumann and Harvard architectures using two popular microcontroller families: STM32 and Arduino AVR.

**1. [Arduino](https://www.ampheo.com/c/development-board-arduino) AVR (Harvard Architecture)**
**Example: [ATmega328P](https://www.ampheo.com/search/ATmega328P) (used in [Arduino Uno](https://www.ampheo.com/product/a000046-25542493))**
* Architecture: Harvard
* Code (Program) Memory: Flash memory (e.g. 32 KB)
* Data Memory: SRAM (e.g. 2 KB)
* Instruction and Data buses: separate
* Instruction word size: 16 bits
* Data word size: 8 bits

**Implications:**
* Can fetch instructions and read/write data at the same time
* Efficient for simple embedded control tasks
* Code and data stored in different physical memory regions

**Example Behavior:**
When you use pgm_read_byte() to read a constant from flash memory, it’s because code and data memory are separate—normal pointers can’t access program memory directly.

**2. [STM32](https://www.ampheo.com/search/STM32) (Von Neumann-Like Architecture)**
**Example: [STM32F103](https://www.ampheo.com/search/STM32F103) (Blue Pill) or [STM32F4](https://www.ampheo.com/search/STM32F4) series**
* Architecture: Modified Harvard / Von Neumann-like
* Unified memory space using bus matrix
* Code and data often reside in the same address space
* Uses caching and pre-fetch to overcome bottlenecks
* Instruction and data fetches may share buses at certain points

**Implications:**
* Simpler software model (code and data are addressable with standard pointers)
* Easier to use C/C++ pointers on both RAM and Flash
* Slight performance tradeoff compared to pure Harvard for some real-time tasks

**Side-by-Side Comparison:**

**Summary:**
* [Arduino](https://www.onzuu.com/manufacturer/arduino) AVR: True Harvard architecture — best for predictable, low-power applications with strict timing.
* [STM32](https://www.ampheoelec.de/search/STM32): Von Neumann-like, more flexible and efficient in software, better for multitasking, RTOS, and high-level features.