IoT Chipsets: Microcontrollers and Embedded Processors Explained

I will be frank enough to say I had thought that all smart devices simply had little computers inside them. At this point I attempted to develop a temperature sensor in my basement, and found myself ignorant on what this chip was. ESP32? STM32? Nordic something? The hole was down into rabbitland.

This is what I came to know after weeks and weeks of trying out boards, burning batteries at an unnecessarily high rate and eventually getting it correct. In case you want to know why your fitness tracker can last weeks or how your smart doorbell can remain connected, you are the one.

MCU vs Microprocessor: What’s Actually Inside Your Gadgets?

Thing that interested me first: do not microcontrollers and microprocessors mean the same thing? Nope.

Microcontroller (MCU): Picture of them as a complete system in a miniature. Microprocessor, memory, disk, over a single chip. It is programmed with the sole aim of doing one particular task very well such as sensor reading or motor control. At low power, the game is called low power.

Microprocessor: It is somewhat a general-purpose brain. It has a lot of powers, though external memory and support chips are required. It is a microprocessor based laptop. Your smart lightbulb? Definitely an MCU.

When I was selecting chips to use in a sensor in the basement, I soon realized that microprocessors were excessive – they would run out of battery in days. MCUs have the capability to execute over months, and even years, using a coin cell.

ARM Cortex-M: The Default Choice

ARM cortex-M processors are used by approximately two-thirds of IoT developers. Why? They have balanced the performance and power consumption. Cortex-M family is ranging between superpart (M0+) and surprisingly powerful (M85).

My initial board was M4 based – possesses sufficient punch to do real-time processing without destroying the battery. It has a mature ecosystem, that is, there are tens of code samples and community support when you find yourself into a jam at 2 am and cannot figure out why your sensor is not booting correctly.

Popular IoT Chipsets I’ve Actually Used (H2)

Having exhausted every false step, the following actually worked:

STM32 Series

The ones manufactured by STMicroelectronics are ubiquitous in industry internet of things. I have chosen an STM32L4 board to make a battery project, which is why this is an L (low power) board and they were not joking around. The current less than 1microamp.

The downside? Increased learning curve compared to others. The tooling (STM32CubeIDE) is not a simple one, but it requires time to learn its ropes.

ESP32: The WiFi Champion

This Espressif model became my favorite to use in any application that required WiFi. Onboard WiFi and Bluetooth Low Energy at an approximate of $5? That’s wild. I have both been able to use ESP32 boards in projects up to smart plant monitors and a homemade air quality sensor.

The catch: WiFi eats power. You will hardly have years of battery life with this one unless you are extremely smart about sleep modes. However, ESP32 is difficult to contest when it is plugged or charged on a regular basis. Besides, prototyping is done quickly due to Arduino compatibility.

Nordic nRF52840: Bluetooth Expert.

Nordic was referred to everyone when I required Bluetooth in a wearable project. The nRF52840 is, in fact, the best when it comes to BLE device models. My battery life on a small lithium cell tapped was 6 months free on a small lithium cell, which is much better than my ESP32 attempts.

The documentation of Nordic is great and their power profiler tool in fact made me identify where my design was taking up too much energy. Begin here in case you are constructing something that goes with a phone.

RisC-V: The Open Source Alternative

I will not lie, I have just recently started working with RISC-V chip. They are making inroads since they do not require any licensing fee as ARM does. The ESP32-C3 has a RISC-V core and can complete most tasks as effectively as the older ESP32 can.

RISC-V is also in its early years, though in case you fear being locked-in with a vendor, or simply use open-source technology, it is worth following. The competitive performance is here and the tool is getting improved within a short period of time.

Power Management: Why Your Smartwatch Lasts a Week

It was the most difficult part that I had to master. Sensors The initial sensor project failed within 3 days. I was taught the lesson the hard way that active power consumption hardly matters but rather it is sleep modes.

Sleep Modes and Wake-on-Interrupt

Current MCUs have a number of sleep states. The ESP32 is founded to consume around 10 microamps of deep sleep. Active WiFi? More like 200 milliamps. That’s a 20,000x difference.

The secret: you can sleep as much as possible and only in a need to wake up. I configure my sensors to go to sleep until an interrupt is generated – such as a button press, or a timer. The MCU brings himself to, works in milliseconds, and goes back to sleep.

This is what was eye opening: a computer that takes up 99 percent sleep time and 1 percent active use can last weeks. Same device always active? Days.

Battery Life Reality Check

Power management through careful attention of power:

• Simple sensors 2-5 years on a coin cell
• Bluetooth wearables: 6-12 months
• WiFi devices Weeks to months (or simply plug in)

I have a temperature sensor that is 18 months and still operating using the starting battery of CR2032. It spins after every 10 minutes where it reads the sensor and goes to sleep. That’s it.

Energy Harvesting: The Future?

Other more recent chips have the ability to pulse power, such as solar energy, vibration energy, and heat energy. I experimented with a solar system to use in an outdoor sensor. Performs excellently during summer, sketchy during winter. The technology exists but is yet to be used in the majority of applications.

Connectivity: Picking the Right Wireless Tech

This is where things get fun. The IoT chips of today squeeze in various wireless choices and wrong selection translates to a rebuild of the entire system.

WiFi: Fast but Hungry

The WiFi provides speed and availability of the current infrastructure. Every home has it. But power-wise, it’s brutal. I do not use WiFi on any devices that are plugged in or charged regularly, smart speakers, security cameras, thermostats.

For battery devices? Unless you are very particular about scheduling your sleep, you might just skip it.

Bluetooth/BLE to Wearables.

Bluetooth Low energy transformed it. The experiments with my fitness tracker came to reality when I switched to BLE now I am getting weeks of battery life. BLE suits phone-paired devices over a distance of between 30 feet.

Normal Bluetooth (Classic) is much more power consuming. Stick with BLE for IoT.

LoRaWAN of Long-range Sensors.

I was doing LoRa testing concerning a farm monitor. It transmitted data more than 2 miles with a line-of-sight with minimally powered data transmission. Seized by the agricultural sensors or the remote monitoring that cannot realize WiFi connection.

The compromise: data rates of extreme low rate. You are not streaming video, you are sending little packets that you send rarely.

Cellular: NB-IoT and LTE-M

Where it is required that projects can operate everywhere, cellular IoT chips such as those in support of either NB-IoT or LTE-M are rational. I have never personally deployed them (carrier costs are high), but it would work well in tracking vehicles or in use in an industry.

The 5G IoT chips are new and in their infancy. The majority of usages do not yet require such bandwidth.

Edge AI: Localizer of Smart Features.

This astonished me: the modern MCUs are able to execute machine learning models on-board. No cloud needed.

TinyML Frameworks

Personally, I have tried out TensorFlow Lite on Microcontrollers dynamical machine (cortex-M4). A basic model of detecting a keyword, such as yes or no, was trained and this also ran completely on-the-job. Less than one hundred milliseconds to respond, hardly any power cost.

Using TinyML is ideal in voice recognition, gesture detection, or visual inspection in sensors in the industrial context. The models will be shrunk to kilobytes and not megabytes.

Why This Matters

Running AI locally means:

• There was no need of internet (privacy win).
• Real-time response (no latency in cloud)
• Weak power (wireless transmission is costly)

Since it is voice-operated, my LED project is quicker than cloud-based applications and is not online. That’s pretty cool for a $10 chip.

Practical Implementations I have Experienced in the Real World

Smart Home Devices

ESP32 dominates here. I have created intelligent light controls, temperature sensors and I even made a garage door monitor. The wireless WiFi and low cost hardware will enable rapid prototyping.

Fitness Trackers and Wearables

Most commercial fitness trackers are operated by Nordic chips. I created a step counter that includes an accelerometer and an nRF52, which entailed numerous hours of fine-tuning the battery life, but the final product had a battery life of 8 months.

Industrial Sensors

STM32 chip deals with factory automation. I also visited a plant with hundreds of vibration sensors in the machines, which were all based on STM32 and predicted which machines would need some maintenance. They have been operating on the same batteries over the years.

Connected Vehicles

This space is heating up fast. Although fully autonomous driving systems require more diligent chips (Big Little Autonomous Driving again, our guide to Automotive Driving Chipsets), vehicle monitoring and diagnostics Simple applications only require regular IoT MCUs.

Agricultural Monitoring

It provided an opportunity to install soil moisture sensors via a small farm with the help of LoRa connectivity. There are sensors distributed on 50 acres that are connected to a central gateway. Due to ultra-low-power design, battery life is 3+ years.

What I would Advise a Person Starting Today

And you are leaping to the world of IoT development:

ESP32 is a good starting point to have wifi and achieve fast results. The community is enormous, there are examples all around and it is inexpensive enough to experiment without undue worry.

Go Nordic (nRF52) when energy conservation and battery are all that are needed, and Bluetooth is essential in your design. This consumes more time to learn but it is worth it with the power efficiency.
STM32 is good when used in an industry or in giving real time performance. The professional ecosystem is really there and so is the learning curve.

Monitor the RISC-V as it grows – more so, when the license price is relevant at scale.

And honestly? It helps a lot to know about semiconductor chipsets, in general. A broader view of the silicon can be found in our Complete Guide to Semiconductor Chipsets in case you want to know how it all works out.

The IoT chip environment continues to change. The edge AI is improving, the energy is getting less, and the connection possibilities are continually created. However, the principles, such as power, performance, and connectivity remain the same.

I am still a beginner, still dropping stuff, still devising more effective ways of coerced months out of a battery. That’s half the fun.