Why Your Smartphone Feels Warm In Your Hand

Pick up your phone after a ten-minute video call over 5G and the back panel is noticeably warm. Play a graphically demanding game for twenty minutes and the aluminium frame gets uncomfortable. Charge it with a 45 W adapter while using navigation and you might see the phone itself warn you that it needs to cool down. None of this is a defect. It is the inevitable consequence of running a system-on-chip that rivals a 2015 laptop processor in computational throughput, inside a sealed enclosure roughly 7 mm thick, with no moving parts and no deliberate airflow.

A modern flagship phone dissipates between 3 and 10 watts under load, with peaks above 12 watts when the cellular radio, display, SoC, and charging circuitry are all active simultaneously. That heat has to go somewhere. There is no fan, no heat sink with fins, no vent. The only path out is conduction through the phone's internal structure and radiation plus convection from the outer surfaces to the surrounding air and to your hand. The warmth you feel is the endpoint of a thermal management chain that starts at a transistor junction running at over 100 degrees Celsius and ends at a glass or aluminium surface held, if the engineers did their job, below 44 degrees Celsius.

This article traces that chain from beginning to end: where the heat comes from, how it spreads, what limits the phone's sustained performance, and why the engineering trade-offs are so tightly constrained that a millimetre of extra thickness can meaningfully change a phone's benchmark scores.

Where The Heat Comes From

A smartphone contains several major heat sources, and understanding their relative contributions matters because they do not all turn on at once.

The SoC is the dominant source. A modern mobile SoC (Qualcomm Snapdragon 8 Gen 3, MediaTek Dimensity 9300, Apple A17 Pro) integrates CPU cores, a GPU, a neural processing unit, an image signal processor, a video codec, a display controller, and in most cases a cellular modem, all on a single die fabricated on a 3 nm or 4 nm process. Under sustained CPU load (all performance cores active, running integer or floating-point workloads), the SoC alone can pull 8 to 10 watts. Under GPU load (high-fidelity gaming at 60 fps), GPU power consumption sits around 5 to 7 watts with the CPU adding another 2 to 3 watts on top.

Measurements from Anandtech and Notebookcheck on flagship 2024 devices show the following typical SoC power draws:

These numbers should be startling. A Snapdragon 8 Gen 3 under full multi-core load burns more power than many ultrabook processors from five years ago, yet it sits inside a passively cooled enclosure that weighs 200 grams.

The cellular modem is the second largest contributor and the least visible one. A 4G LTE modem transmitting at full power draws around 1 to 2 watts. A 5G sub-6 GHz modem adds about 1 watt on top of that. And 5G mmWave, which requires a phased-array antenna module with multiple power amplifiers, can add 2 to 3 watts of additional radio power. The power amplifier in a mmWave front-end module is notoriously inefficient: typical PA efficiency for mmWave is 20 to 30 percent, which means that for every watt of radiated RF power, you dissipate 2 to 3 watts as heat in the PA itself. This is why phones get noticeably warmer during long video calls on 5G than on Wi-Fi, and why some phones silently fall back from 5G to 4G when thermal limits are approached.

The display is a constant, always-on heat source. An OLED panel at moderate brightness (around 200 nits indoors) draws about 0.8 to 1.2 watts on a typical 6.7-inch phone screen. Increase the brightness to outdoor-readable levels (1500 to 2000 nits, which modern flagship panels support) and display power climbs to 2 to 3 watts. LCD panels are often slightly worse at high brightness because the backlight must illuminate all pixels regardless of content, while OLED only powers the lit pixels. Both technologies generate heat, and the display is physically the largest component in the phone, so its thermal contribution is spread across the entire front surface.

The charging circuitry produces significant heat during fast charging. A phone accepting 45 watts from a USB-C PD charger does not deliver all 45 watts to the battery. The charging IC (the buck converter that steps down the input voltage to the battery voltage) has an efficiency of 90 to 95 percent at best, which means 2 to 5 watts of heat dissipation in the charging IC alone. On top of that, the battery itself generates heat from its internal resistance as current flows through it. At high charging currents (4 to 5 amps on a single cell, or distributed across dual cells in some designs), the battery's internal resistance of 50 to 80 milliohms dissipates I squared R watts: at 4 A, that is about 0.6 to 1.3 watts of heat inside the battery itself. Sum the charging IC losses and the battery losses and fast charging can add 3 to 6 watts of thermal load to the system. This is why most phones throttle their charging rate as the battery warms up, and why the last 20 percent of a fast charge is not actually fast.

Miscellaneous sources include the Wi-Fi radio (0.5 to 1 W during active transfer), the LPDDR5X memory (0.5 to 1.5 W under load), the NAND flash controller (0.3 to 0.8 W during sustained writes), and GPS reception (0.1 to 0.3 W). Individually these are small, but they all add to the thermal budget and they all generate heat inside the same sealed enclosure.

When everything is active simultaneously (gaming on 5G with the screen at full brightness while the phone is charging), the total power dissipation can exceed 15 watts. That is an extraordinary amount of heat for a passively cooled device.

The Skin Temperature Limit: Mobile Thermal Design

Desktop and laptop processors have a Thermal Design Power (TDP) rating: the Intel Core i7 in a laptop might be rated at 28 watts, meaning the cooling system is designed to handle 28 watts of sustained heat dissipation. Mobile phones do not use TDP. Instead, they use a skin temperature limit, sometimes called a thermal design point.

The reason is human ergonomics. A phone is a device you hold against your skin for extended periods. The outer surface temperature matters more than the silicon junction temperature because it directly affects user comfort and safety. The industry consensus, informed by IEC 62368-1 safety standards and OEM testing, is that phone surface temperature should not exceed approximately 40 to 44 degrees Celsius during sustained use. Above 44 degrees Celsius, the device feels uncomfortably hot. Above 48 degrees Celsius, there is a risk of mild thermal discomfort with prolonged contact. Above 60 degrees Celsius (which phones should never reach in normal operation), there is a burn risk.

The skin temperature limit defines the entire thermal architecture of a phone. The question is not "how much power can the silicon dissipate?" but rather "how much power can the phone dissipate while keeping the outer surface below 43 degrees Celsius in a 25 degree Celsius ambient environment?" The answer, for a typical modern flagship, is roughly 4 to 6 watts sustained. That is the phone's true thermal budget: not the peak SoC power, not the charging power, but the maximum continuous heat dissipation that keeps the skin temperature within bounds.

This is a severe constraint. The SoC can burst to 10 watts or more, but it cannot sustain that level. Within 2 to 5 minutes of sustained full load, the thermal management system will begin throttling clocks to bring power consumption down to the 4 to 6 watt sustainable range. The phone's sustained performance is therefore not defined by its peak silicon capability, but by its thermal envelope.

To put this in perspective with real numbers: the thermal resistance from the SoC junction to the phone's outer surface is typically 8 to 12 K/W (degrees Celsius per watt) depending on the phone's construction. If the ambient temperature is 25 degrees Celsius and the skin temperature limit is 43 degrees Celsius, the available temperature delta is 18 K. Dividing 18 K by the thermal resistance of 10 K/W gives a sustained power budget of 1.8 watts from the SoC alone through that path. Fortunately, the phone spreads heat across its entire surface area and the effective thermal resistance to the environment is much lower when the whole enclosure participates in heat dissipation, which is why the practical sustained budget reaches 4 to 6 watts total rather than 1.8 watts. Heat spreading is therefore not a luxury; it is essential to making the phone usable at all.

The Thermal Path: From Junction To Skin

Heat generated at the SoC die must travel through a series of materials before it reaches the phone's outer surface. Each material in the path has a thermal resistance, and the total resistance of the chain determines how large the temperature difference is between the die and the skin. Understanding this chain explains why phone design decisions that seem cosmetic (glass back versus plastic back, aluminium frame versus stainless steel) have real performance consequences.

Step 1: Die to package. The SoC die is a thin piece of silicon, typically 0.3 to 0.5 mm thick, mounted on an organic substrate via flip-chip solder bumps. The silicon itself has excellent thermal conductivity (about 150 W/mK), so thermal resistance within the die is low. The solder bumps and underfill between the die and the substrate add some resistance, but this interface is designed to be efficient.

Step 2: Package to thermal interface material. Unlike a desktop CPU, a phone SoC has no metal heat spreader lid. The die is either bare or covered with a thin thermal interface material (TIM) that contacts the phone's internal heat-spreading structure. This TIM is typically a thermal pad or thermal paste with a conductivity of 3 to 8 W/mK, vastly worse than the copper or aluminium used in desktop cooling. The TIM is a significant bottleneck in the thermal path.

Step 3: Heat spreader. Modern flagship phones use a heat-spreading layer immediately above the SoC. This might be a graphite sheet, a vapour chamber, or a copper heat pipe. Its job is to take the concentrated heat from the small SoC die (roughly 100 mm squared) and spread it over a much larger area of the phone's internal structure. Without this spreading, the area directly above the SoC would be a painful hot spot while the rest of the phone remained cool, which is both uncomfortable and thermally inefficient.

Step 4: Metal midframe. The phone's structural midframe is typically cast or machined aluminium (thermal conductivity: 205 W/mK) or stainless steel (thermal conductivity: 16 W/mK). Aluminium is far superior as a heat conductor, which is one reason why aluminium-framed phones tend to have better sustained performance than stainless-steel-framed ones, all else being equal. The midframe acts as both a structural member and a heat spreader, conducting heat laterally from the centre of the phone toward the edges.

Step 5: Battery as thermal mass. The battery occupies 40 to 50 percent of the phone's internal volume. It is a large thermal mass (lithium-ion cells have a specific heat capacity of about 1.0 to 1.1 J/gK, and a 5000 mAh battery weighs about 60 grams). During short bursts of high power, the battery absorbs heat and buffers the temperature rise. During sustained load, it reaches thermal equilibrium and stops absorbing. The battery's thermal role is therefore most significant in the first few minutes of a heavy workload, buying time before the thermal management system needs to intervene.

Step 6: Enclosure to environment. The outer shell of the phone (glass back, aluminium frame, or plastic back) is the final interface with the outside world. Heat leaves the enclosure through three mechanisms: conduction to whatever the phone is touching (your hand, a table, a wireless charger), convection to the surrounding air, and radiation. The relative contributions depend on the situation. When you hold the phone, conduction to your hand is the dominant path. When the phone is lying on a table, convection and radiation dominate from the exposed surface.

Glass backs (thermal conductivity: about 1 W/mK) are terrible thermal conductors. They do not spread heat laterally at all, which means the surface temperature directly above the SoC is much higher than the surface temperature at the edges. Aluminium (205 W/mK) spreads heat efficiently across the whole surface. Frosted glass with a thin metal layer behind it is a compromise that many manufacturers use. The material choice for the back panel is a genuine thermal engineering decision, not just an aesthetic one.

Modelling The Phone As An RC Circuit

Thermal engineers routinely model heat flow using electrical analogies, because the mathematics is identical. Temperature maps to voltage, heat flow (in watts) maps to current, thermal resistance maps to electrical resistance, and thermal capacitance (the ability to store heat) maps to electrical capacitance. A phone's thermal behaviour can be approximated as a network of resistors and capacitors, and this model predicts both the steady-state temperatures and the transient response to sudden changes in power.

The simplest useful model has three nodes:

1. Junction (T_j): the SoC die temperature. This is the hottest point in the system.
2. Skin (T_s): the outer surface temperature. This is what the user feels and what the thermal policy regulates.
3. Ambient (T_a): the surrounding air temperature. This is the heat sink.

Between the junction and the skin, there is a thermal resistance R_js (typically 8 to 12 K/W) and a thermal capacitance C_j (representing the thermal mass of the SoC, the heat spreader, and the battery, roughly 15 to 30 J/K). Between the skin and the ambient, there is a thermal resistance R_sa (typically 15 to 25 K/W for natural convection from the phone's surface area) and a thermal capacitance C_s (representing the thermal mass of the enclosure, roughly 5 to 10 J/K).

In steady state, the capacitors are irrelevant and the temperature rise is purely resistive:

T_j = T_a + P * (R_js + R_sa)
T_s = T_a + P * R_sa

Where P is the total power dissipation. For a phone dissipating 5 watts with R_js = 10 K/W and R_sa = 20 K/W, and an ambient temperature of 25 degrees Celsius:

T_j = 25 + 5 * (10 + 20) = 175 °C   (too hot, the die would be damaged)
T_s = 25 + 5 * 20 = 125 °C           (would burn the user)

These numbers are obviously unacceptable, and they reveal why the model is oversimplified. In reality, the phone does not dissipate 5 watts through a single thermal path with 20 K/W to ambient. Heat spreads across the entire phone surface (which might have an area of 200 cm squared, including both sides), and the effective R_sa is much lower because the total surface area participating in convection and radiation is large. A more realistic effective R_sa for the whole phone is about 3 to 5 K/W, which gives:

T_s = 25 + 5 * 4 = 45 °C     (warm but tolerable)
T_j = 25 + 5 * (10 + 4) = 95 °C  (within the silicon's safe range)

This is much closer to measured reality. A phone dissipating 5 watts sustained reaches a skin temperature of about 42 to 46 degrees Celsius and a junction temperature of 90 to 100 degrees Celsius. The numbers work out precisely because the heat-spreading system distributes the thermal load across the entire enclosure.

The transient response is where the capacitors matter. When you launch a demanding game, the SoC ramps from 1 watt idle to 8 watts immediately. The junction temperature rises quickly (the SoC's own thermal mass is small, maybe 1 to 2 J/K). The skin temperature rises slowly (the battery and enclosure have large thermal mass, absorbing heat). The time constant of the junction node is:

tau_j = R_js * C_j = 10 * 20 = 200 seconds

The time constant of the skin node is:

tau_s = R_sa * C_s = 4 * 8 = 32 seconds

But the dominant time constant that the user perceives (how long until the phone feels warm) is governed by the total thermal mass between the SoC and the skin, which is dominated by the battery. In practice, it takes 3 to 5 minutes for the phone surface to reach its steady-state temperature after a sudden increase in power. This is why benchmark scores start high and decline over time: the silicon can run at full speed for several minutes before the skin temperature reaches the limit and throttling kicks in.

Heat Spreading Technologies

The gap between the SoC die (roughly 100 mm squared, about the size of a fingernail) and the phone's total surface area (roughly 20,000 mm squared, both sides combined) is a factor of 200. If the heat stayed concentrated above the SoC, the local surface temperature would be extreme while the rest of the phone stayed cool. The entire point of the heat-spreading system is to distribute that concentrated heat source across the largest possible area, reducing the peak surface temperature and making the thermal budget usable.

Graphite sheets are the simplest and most universal solution. A thin sheet of pyrolytic graphite (0.025 to 0.1 mm thick) has an in-plane thermal conductivity of 1000 to 1500 W/mK, which is 5 to 7 times better than copper. The through-plane conductivity is much lower (about 5 to 10 W/mK), so graphite is excellent at spreading heat laterally but poor at conducting it vertically. Phones typically have one or more graphite sheets layered between the SoC and the back panel, covering a large fraction of the phone's footprint. Graphite is light, thin, inexpensive, and reliable, which is why even budget phones use it.

Vapour chambers are the premium solution used in flagship phones (Samsung Galaxy S24 Ultra, gaming phones like the ASUS ROG Phone 8, and many Chinese flagship devices). A vapour chamber is a flat, sealed copper enclosure (typically 0.3 to 0.6 mm thick) containing a small amount of working fluid (usually water) and a wick structure. When the SoC heats one spot on the chamber, the fluid evaporates, the vapour spreads rapidly across the chamber, condenses on cooler areas, and the condensed liquid wicks back to the hot spot. This two-phase cycle moves heat much more efficiently than conduction alone.

The effective thermal conductivity of a vapour chamber can exceed 10,000 W/mK in the spreading plane, which makes it far more effective than even graphite for distributing a concentrated heat source. The practical benefit is a reduction in the peak skin temperature hot spot of 3 to 5 degrees Celsius compared to graphite alone, which translates directly into either a higher sustained power budget or a more comfortable surface temperature.

The size of the vapour chamber matters enormously. Early phone vapour chambers were small (around 400 mm squared), which limited their effectiveness. Modern flagships use vapour chambers of 2000 to 4000 mm squared, sometimes covering most of the phone's internal area. Samsung's Galaxy S24 Ultra uses a vapour chamber of roughly 3500 mm squared, and ASUS's ROG Phone 8 Pro goes larger still.

Copper heat pipes are an older solution, less common in current flagships but still used in some mid-range devices. A heat pipe is a sealed copper tube (flattened to fit inside a phone) with a wick and working fluid. It works on the same principle as a vapour chamber but spreads heat along one axis rather than across a plane. Heat pipes are less effective for area spreading but simpler to manufacture.

Thermal pads and thermal paste fill the gaps between components and the heat-spreading layers. Their conductivity (3 to 8 W/mK for typical pads) is far lower than the spreaders themselves, and they are often the bottleneck in the thermal path. Thinner pads with higher conductivity materials (boron nitride or graphene-enhanced composites) are an area of active development.

DVFS: The Primary Thermal Control Mechanism

Dynamic Voltage and Frequency Scaling (DVFS) is the single most important thermal management tool in a phone. It is the mechanism by which the SoC reduces its power consumption in response to rising temperature, and it is the reason why benchmark scores decline over time, why phones feel slower when they are hot, and why the same phone performs differently in Athens in August and Helsinki in January.

The relationship between voltage, frequency, and power in a CMOS circuit is governed by:

P_dynamic = C * V^2 * f * alpha

Where C is the total switched capacitance, V is the supply voltage, f is the clock frequency, and alpha is the activity factor (the fraction of transistors switching per cycle). The key insight is the V squared term. Power scales with the square of the voltage, and higher clock frequencies require higher voltages to maintain signal integrity. Reducing the frequency from 3.3 GHz to 2.5 GHz on a Cortex-X4 core does not just reduce the frequency by 24 percent; it also allows the voltage to drop from perhaps 1.05 V to 0.85 V, and the power reduction is:

P_ratio = (0.85/1.05)^2 * (2.5/3.3) = 0.655 * 0.758 = 0.50

Halving the power by giving up 24 percent of the clock speed. This is an excellent trade-off when the phone is thermally limited, and it is the reason DVFS is so effective.

The DVFS governor is a piece of firmware (or a kernel driver, depending on the platform) that monitors several thermal sensors on the SoC die, compares their readings to a set of temperature thresholds, and adjusts the operating point (voltage and frequency) accordingly. The typical control loop works as follows:

1. Thermal sensors on the die report junction temperature every 1 to 10 milliseconds.
2. The governor compares the temperature to a set of trip points, typically defined in the device tree or ACPI tables.
3. Below the first trip point (say 85 degrees Celsius junction temperature), the governor allows full performance.
4. Above the first trip point, the governor begins stepping down frequency. The step size and rate depend on the policy, but a common approach is to reduce the maximum allowed frequency by one operating performance point (OPP) per trip, where each OPP represents a discrete voltage-frequency pair.
5. If the temperature continues to rise to the critical trip point (say 110 degrees Celsius), the governor forces the lowest operating point or shuts down cores entirely.
6. As the temperature drops, the governor raises the allowed frequency, with hysteresis to prevent oscillation.

On Android phones using the Linux kernel, the thermal framework exposes this as a cooling device bound to a thermal zone. You can inspect the current state on a rooted device:

# Show thermal zones and their temperatures
for tz in /sys/class/thermal/thermal_zone*; do
    echo "$(cat $tz/type): $(cat $tz/temp)"
done
 
# Show cooling device state (0 = no throttling, higher = more throttled)
for cd in /sys/class/thermal/cooling_device*; do
    echo "$(cat $cd/type): state=$(cat $cd/cur_state) max=$(cat $cd/max_state)"
done

On a Snapdragon 8 Gen 3 device, you might see thermal zones for cpu-0-0-0 (individual core), gpuss-0 (GPU), nsp (NPU), and mdmss (modem). Each has its own trip points and cooling policies, and the governor manages them independently.

Apple's approach is similar in principle but completely proprietary. The A17 Pro's thermal management is handled by firmware running on the SoC's always-on processor, and the trip points and policies are not exposed to userspace. The observable effect is the same: sustained benchmarks show declining scores as the die heats up.

big.LITTLE And DynamIQ: Heterogeneous Cores As Thermal Strategy

The heterogeneous core architecture used in every modern phone SoC (ARM's big.LITTLE, now DynamIQ) is partly a thermal story. A Snapdragon 8 Gen 3 has:

• 1 Cortex-X4 "prime" core at up to 3.3 GHz
• 3 Cortex-A720 "performance" cores at up to 3.15 GHz
• 4 Cortex-A520 "efficiency" cores at up to 2.3 GHz

The Dimensity 9300 went further with an all-big-core design:

• 1 Cortex-X4 at 3.25 GHz
• 3 Cortex-X4 at 2.85 GHz
• 4 Cortex-A720 at 2.0 GHz

Apple's A17 Pro uses:

• 2 high-performance cores at 3.78 GHz
• 4 high-efficiency cores at 2.11 GHz

The thermal rationale is straightforward. A Cortex-X4 core at 3.3 GHz burns about 5 watts by itself. A Cortex-A520 efficiency core at 2.3 GHz burns about 0.3 watts. For workloads that do not need peak single-thread performance (background sync, UI rendering, messaging), running them on efficiency cores uses one-tenth the power and generates one-tenth the heat. The scheduler (on Android, the Energy Aware Scheduler or EAS) chooses which core to use based on the task's utilisation history, the current thermal state, and the energy cost of each operating point.

This is why your phone does not feel warm when you are reading email or scrolling social media: those tasks run almost entirely on efficiency cores. The performance cores only fire during visually complex UI transitions, app launches, and sustained compute workloads. And even then, the scheduler tries to spread work across cores to avoid concentrating heat in one part of the die.

The Dimensity 9300's all-big-core approach is an interesting counterpoint. MediaTek bet that four "medium" cores (A720 at 2.0 GHz, roughly 1 watt each) could handle light tasks efficiently enough, while gaining significantly more performance when all eight cores are loaded. The thermal consequence is a chip that runs hotter under multi-core load, and early reviews noted more aggressive throttling on Dimensity 9300 devices compared to Snapdragon 8 Gen 3 devices in the same benchmarks. Whether this trade-off is worthwhile depends on how often the user actually needs all-core performance versus efficiency-core idle.

Thermal Throttling In Practice

The gap between a phone's peak performance and its sustained performance is one of the most misunderstood aspects of mobile hardware. Reviewers who run a single-iteration benchmark see the peak. Users who play a demanding game for thirty minutes experience the sustained level, which can be 30 to 50 percent lower.

The most commonly cited tool for measuring this is the CPU Throttling Test, which runs a sustained multi-thread workload and reports performance as a percentage of the initial score over time. Typical results for flagships in a 25 degree Celsius room:

Snapdragon 8 Gen 3 (Samsung Galaxy S24 Ultra):

• 0 to 3 minutes: 100 percent performance
• 3 to 5 minutes: drops to 85 percent
• 5 to 10 minutes: stabilises around 70 to 75 percent
• 15 to 30 minutes: settles at 65 to 70 percent

Apple A17 Pro (iPhone 15 Pro):

• 0 to 3 minutes: 100 percent performance
• 3 to 5 minutes: drops to 80 percent
• 5 to 10 minutes: stabilises around 70 percent
• 15 to 30 minutes: settles at 65 to 70 percent

Dimensity 9300 (vivo X100 Pro):

• 0 to 3 minutes: 100 percent performance
• 3 to 5 minutes: drops to 75 percent
• 5 to 10 minutes: stabilises around 60 to 65 percent
• 15 to 30 minutes: settles at 55 to 65 percent

These numbers vary significantly between devices using the same SoC, because the phone manufacturer's cooling solution and thermal policy determine the sustained performance. A Samsung Galaxy S24 Ultra with its large vapour chamber sustains higher performance than a smaller, thinner phone using the same Snapdragon 8 Gen 3.

Apple's approach is notable for being more aggressive with early throttling. The A17 Pro drops to its sustained level faster but holds it more consistently, resulting in a flatter performance curve after the initial drop. Qualcomm and MediaTek devices tend to have a more gradual decline with occasional further drops as the battery and frame heat up.

The GPU story is similar. In 3DMark Wildlife Extreme Stress Test (which loops a graphics benchmark for 20 minutes), typical throttling results show:

• Best loop score: 100 percent
• Worst loop score (after 15 to 20 minutes): 50 to 65 percent of the best
• Stability rating: 50 to 70 percent for most flagships

A stability rating of 60 percent means the phone's sustained GPU performance is 60 percent of its peak, which is the number that actually determines gaming frame rates after the first few minutes.

The 5G Modem: A Hidden Heat Source

The cellular modem deserves its own discussion because its thermal contribution is significant and poorly understood by most users.

A 5G NR modem consists of a digital baseband processor and one or more radio frequency front-end modules (RFFEMs). The baseband handles channel coding, MIMO processing, and protocol operations. The RFFEM handles power amplification, filtering, and antenna switching. Both produce heat.

In sub-6 GHz operation (the most common 5G deployment in Europe, using bands around 3.5 GHz), the modem's total power draw during active data transfer is about 2 to 3 watts. This is roughly 1 watt more than an equivalent 4G LTE connection at the same data rate, because 5G NR uses wider channel bandwidths (100 MHz versus 20 MHz for LTE), more MIMO layers, and higher-order modulation, all of which increase the baseband's processing load and the PA's output power.

In mmWave operation (deployed in limited urban areas, using frequencies above 24 GHz), the situation is worse. mmWave requires a phased-array antenna module with multiple small power amplifiers, each driving one element of the array. The total radiated power might be modest (200 to 400 mW), but the PA efficiency at 28 or 39 GHz is only 20 to 30 percent, which means the PAs alone dissipate 1 to 2 watts. Add the beamforming controller and the wideband baseband processing, and mmWave operation can add 3 to 4 watts of total system power compared to a 4G connection.

Qualcomm's Snapdragon X75 modem (used in the Snapdragon 8 Gen 3) integrates the 5G modem on the same die as the application processor. This has a thermal advantage (the heat is spread across one die and managed by one thermal system) but also means the modem's heat directly competes with the CPU and GPU for the same thermal budget. During a video call on 5G, the modem might be consuming 2.5 watts while the video codec uses 1 watt and the display uses 1.5 watts, leaving only 1 to 2 watts of thermal headroom for the CPU. This is why phones sometimes feel sluggish during 5G video calls: the CPU is being throttled to make room for the modem's thermal contribution.

MediaTek and Samsung (with Exynos) use a similar integrated approach. Apple's transition to its own modem (expected in the iPhone SE 4 and beyond) is partly motivated by thermal efficiency: a purpose-designed modem integrated into Apple's own process technology could potentially reduce modem power and improve the thermal budget available to the CPU and GPU.

Battery Chemistry: The Temperature Ceiling You Cannot Negotiate

The lithium-ion cells in your phone impose a hard thermal constraint that has nothing to do with user comfort or silicon survival. Battery chemistry degrades rapidly at elevated temperatures, and the phone's thermal management system protects the battery as aggressively as it protects the SoC.

The relevant numbers:

• Optimal operating range: 15 to 35 degrees Celsius. Battery degradation rate is minimised in this window.
• Elevated temperature threshold: 40 to 45 degrees Celsius. Degradation rate roughly doubles for every 10 K increase above 25 degrees Celsius. A battery regularly operated at 45 degrees Celsius will lose 20 to 30 percent of its capacity in two years instead of the typical 10 to 15 percent.
• Safety limit: 60 degrees Celsius. Above this temperature, the electrolyte begins to decompose, generating gas. The cell swells, the internal resistance increases, and there is a risk of thermal runaway. Phone battery management systems (BMS) will shut down charging and aggressively throttle the SoC before the cell reaches this temperature.
• Catastrophic failure: above 80 to 100 degrees Celsius (depending on chemistry and state of charge), thermal runaway becomes self-sustaining: the exothermic decomposition of the cathode material generates enough heat to propagate, and the cell can vent flame.

The battery is physically adjacent to the SoC in most phone designs, separated by only a few millimetres and the metal midframe. Heat from the SoC conducts into the battery, raising its temperature. The phone's thermal policy therefore monitors battery temperature alongside junction temperature, and will throttle performance to keep the battery below 40 to 43 degrees Celsius regardless of whether the SoC is within its own thermal limits.

This battery-protective throttling is most noticeable during gaming while charging. The charging circuitry adds heat (3 to 5 watts), the SoC adds heat (5 to 8 watts), and the battery is caught between two heat sources while simultaneously generating internal heat from the charging current. Many phones will reduce the charging rate, throttle the SoC, or both when this combination is detected. Some gaming-focused phones (ASUS ROG Phone, Xiaomi Black Shark) offer a "bypass charging" mode that powers the phone directly from the charger while the game is running, avoiding the battery heating entirely.

Charging speed itself is thermally limited. The fast-charging arms race (65 W, 120 W, 240 W on some Chinese market phones) is constrained by battery temperature rise. Higher charging power means higher current, higher I squared R losses, and faster temperature rise. Manufacturers work around this with dual-cell battery designs (two cells in series, halving the current per cell for a given power level), higher-conductivity tab designs, and elaborate thermal management around the charging IC. But the fundamental limit remains: you cannot push charging current through a lithium-ion cell's internal resistance without generating heat, and the cell's chemistry sets a non-negotiable temperature ceiling.

Environmental Factors: Athens In August vs Helsinki In January

The ambient temperature has a direct and quantifiable effect on phone performance, because the phone's thermal budget is the difference between the skin temperature limit and the ambient temperature.

Consider a phone with a skin temperature limit of 43 degrees Celsius and an effective thermal resistance of 4 K/W from the SoC to the skin. The sustainable power budget is:

P_sustained = (T_skin_max - T_ambient) / R_effective

In Helsinki in January, with an ambient temperature of minus 10 degrees Celsius (or 0 degrees Celsius if the phone is in your pocket):

P_sustained = (43 - 0) / 4 = 10.75 W

In Athens in August, with an ambient temperature of 40 degrees Celsius (easily reached on a sunny terrace):

P_sustained = (43 - 40) / 4 = 0.75 W

The difference is staggering. In cold conditions, the phone has an enormous thermal budget and can sustain near-peak performance almost indefinitely. In hot conditions, the thermal budget nearly vanishes, and the phone must throttle aggressively even for moderate workloads.

This is why phones feel slower in summer. It is not your imagination. A phone sitting on a car dashboard in direct sun in southern Europe, where surface temperatures can reach 50 to 60 degrees Celsius, may throttle to its minimum operating point and display a thermal warning. The phone is not broken; its thermal budget is simply negative at those ambient temperatures (the enclosure is already above the skin temperature limit from solar heating alone, before the SoC generates any heat at all).

The cold extreme has its own problems. Lithium-ion batteries deliver less current at low temperatures because the ionic conductivity of the electrolyte drops. Below 0 degrees Celsius, the battery's internal resistance increases significantly, voltage sag under load becomes severe, and the phone may shut down even with 30 percent charge remaining. Apple and Samsung both implement cold-weather protections that reduce peak current draw (and therefore peak performance) when the battery temperature is below about 5 degrees Celsius.

The practical consequence is that phone performance has a seasonal and geographic dimension that reviewers rarely account for. A benchmark run in an air-conditioned lab at 22 degrees Celsius produces results that a user in Athens in summer or Dubai in any season will never reproduce.

Where The Hot Spots Are

Teardown thermal imaging of phones (using FLIR or similar infrared cameras) consistently reveals the same heat distribution pattern.

With the back panel removed, during sustained CPU load:

The hottest point is directly over the SoC, typically located in the upper third of the phone near the cameras. Junction temperatures of 95 to 105 degrees Celsius produce a visible hot spot on the package surface, with temperatures of 50 to 65 degrees Celsius on the exposed SoC package. The vapour chamber or graphite sheet below the back panel shows a thermal gradient from the SoC outward, with the temperature dropping by 10 to 15 degrees Celsius from the centre to the edges.

The power management IC (PMIC), located near the SoC, is the second hottest component, often reaching 60 to 70 degrees Celsius under load. The PMIC handles voltage regulation for the SoC's multiple power rails and dissipates 1 to 2 watts of its own.

The modem area (if separate from the SoC) or the RF front-end modules show elevated temperatures during active cellular communication. The power amplifiers for the cellular radio can reach 60 to 80 degrees Celsius during transmission.

The charging IC, located near the USB-C port at the bottom of the phone, becomes a visible hot spot during fast charging. Temperatures of 55 to 70 degrees Celsius on the charging IC are common at high charging rates.

From the outside, with a thermal camera pointed at the back of the phone during a sustained gaming session:

The hottest area on the back panel is typically offset from the SoC's actual location by a centimetre or two, because the vapour chamber and graphite sheets redirect heat flow. Peak surface temperatures of 42 to 46 degrees Celsius are common on flagships, with 2 to 4 degrees Celsius difference between the hottest and coolest points on the back panel. On phones with effective heat spreading, this gradient is small (the whole back feels uniformly warm). On phones with poor heat spreading, you can feel a distinct hot spot with your fingertip.

The frame edges of aluminium-framed phones get warm, sometimes warmer than the back panel, because aluminium conducts heat efficiently to the frame. This is actually desirable: the frame has a large surface area exposed to air and acts as a secondary heat sink.

Why Phones Have No Fans

The absence of active cooling in phones is not a technological limitation. It is a deliberate engineering choice driven by five constraints that are each individually sufficient to rule out fans:

Power budget. A small fan capable of moving meaningful airflow (say, 0.5 CFM) consumes 0.3 to 0.5 watts. In a device with a total thermal budget of 4 to 6 watts, spending 10 percent of it on a fan is hard to justify, especially since the fan's benefit is modest in such a thin enclosure with limited airflow paths.

Space. A phone's internal volume is almost entirely occupied by the battery (40 to 50 percent), the display stack (20 to 25 percent), the PCB and SoC (10 to 15 percent), the camera modules (5 to 10 percent), and connectors and structural elements. There is no room for a fan and its air channel without sacrificing battery capacity or making the phone thicker.

Waterproofing. Modern flagships carry IP68 ratings (submersion to 1.5 metres for 30 minutes). A fan requires an air inlet and outlet, which are holes in the enclosure. Waterproofing a phone with air holes is possible but vastly more complex and fragile. Losing IP68 to gain a fan is a trade-off no major manufacturer has been willing to make in a mainline flagship.

Dust and debris. Air intake means dust intake. Phone users carry their devices in pockets full of lint, on sandy beaches, in dusty construction sites. A fan that ingests particles would clog, degrade, and fail. Desktop computers have this problem at a manageable level; a phone-sized device would have it at an intolerable level.

Noise and vibration. A fan small enough for a phone would spin at very high RPM (10,000 or more) to move any air at all, producing a high-pitched whine that would be audible during phone calls and in quiet environments. Any slight imbalance would transmit vibration through the phone's rigid structure.

Some gaming phone accessories (clip-on Peltier coolers, fan cases) exist as aftermarket products. They work (a Peltier cooler can drop the phone's back panel temperature by 10 to 15 degrees Celsius, significantly expanding the thermal budget), but they add bulk, consume their own battery or USB power, and are impractical for everyday use. They exist because the thermal constraints on a passively cooled phone are genuinely severe, and some users are willing to trade portability for sustained gaming performance.

The Thickness Trade-Off

Phone thickness is the single design parameter that most directly affects thermal performance, and the industry trend toward thinner phones has measurably worsened sustained performance.

The relationship is physical. A thicker phone has:

• More internal volume for a larger vapour chamber or more graphite layers.
• A larger battery, which provides more thermal mass to buffer transient loads.
• More physical distance between the SoC and the outer surface, allowing heat to spread over a wider area before reaching the skin.
• More room for innovative cooling structures (thicker vapour chambers are more effective because the vapour has more room to flow).

Consider two hypothetical phones with identical SoCs and identical surface areas, differing only in thickness: 7.5 mm versus 9.5 mm. The thicker phone might have a vapour chamber that is 0.5 mm thick instead of 0.3 mm, a battery that is 5500 mAh instead of 4500 mAh, and an additional graphite layer. The net effect could be a 1 to 2 K/W reduction in thermal resistance and a 30 percent increase in thermal mass, translating to 15 to 25 percent better sustained performance.

This is not hypothetical. Gaming phones (ASUS ROG Phone 8 Pro: 8.9 mm, 232 grams; Redmagic 9 Pro: 8.9 mm, 229 grams) consistently demonstrate better sustained performance than thinner mainstream flagships (Samsung Galaxy S24: 7.6 mm, 167 grams; iPhone 15 Pro: 8.25 mm, 187 grams) using comparable or identical SoCs. The ROG Phone 8 Pro sustains 80 to 85 percent of its peak CPU performance after 30 minutes where the Galaxy S24 sustains 65 to 70 percent.

The industry knows this. Every phone manufacturer's thermal engineering team can tell you exactly how many degrees of thermal headroom they lose per tenth of a millimetre of thickness reduction. But the industrial design teams want thin phones, the marketing teams want thin phones, and consumers pick up phones in shops and judge them partly by how slim they feel. The result is a persistent tension where thermal performance is sacrificed for aesthetics, and the sacrifice shows up as throttling that most users attribute to "the phone being slow" rather than to the phone being thin.

Comparison With Laptop Thermal Design

Comparing a phone to a laptop makes the constraints concrete.

A modern ultrabook with an Intel Core Ultra or AMD Ryzen AI processor has:

• TDP: 15 to 28 watts sustained, with turbo bursts to 45 to 65 watts.
• Cooling: one or two fans, one or two heat pipes, a copper or aluminium heat sink with fins, and vent holes in the chassis.
• Thermal resistance to ambient: roughly 0.5 to 1.5 K/W, achieved through forced air convection over a finned heat sink.
• Enclosure surface area: much larger than a phone, and the keyboard deck and bottom panel both participate in passive dissipation.
• Thickness: 15 to 20 mm, giving ample room for heat pipes and fan assemblies.

A phone has:

• Sustained thermal budget: 4 to 6 watts.
• Cooling: no fans, no fins, no vents. Passive conduction and natural convection only.
• Thermal resistance to ambient: roughly 3 to 5 K/W, limited by the small surface area and absence of forced convection.
• Thickness: 7 to 9 mm.

The laptop can sustain 4 to 6 times the power dissipation of a phone because it has an active cooling system with 3 to 10 times lower thermal resistance to the environment. And yet the silicon in both devices is increasingly comparable. The Cortex-X4 core in a Snapdragon 8 Gen 3 trades blows with a Cortex-A78AE or Cortex-X1 core in a laptop Arm SoC, and Apple's phone and laptop chips share the same core microarchitecture (the A17 Pro's performance core is the same microarchitecture as the M3's performance core).

The implication is that mobile silicon is increasingly capable of laptop-class performance in short bursts, but the thermal envelope forces it back to a fraction of that capability within minutes. The gap between peak and sustained performance in a phone (30 to 50 percent) is far larger than in a well-cooled laptop (typically 10 to 20 percent). This is the unavoidable cost of fitting powerful silicon into a pocketable, sealed, passively cooled device.

Software And The Thermal Budget

Hardware is only half the story. Software determines how much of the thermal budget is spent on useful work versus waste.

Efficient applications are thermally invisible. A well-optimised messaging app uses efficiency cores at low frequency, touches the GPU only for compositing, and keeps the modem in low-power DRX (Discontinuous Reception) mode between message checks. Total incremental power: 0.1 to 0.3 watts. The phone does not warm up.

Inefficient applications burn thermal budget for nothing. Common offenders include:

• Excessive wake locks: apps that prevent the CPU from entering deep sleep states to run background tasks that could be batched. A single app holding a partial wake lock and polling a server every 10 seconds can keep a performance core active continuously, burning 0.5 to 1 watt that contributes nothing the user can see.
• Unthrottled rendering: apps that render UI at the maximum frame rate even when nothing on screen is changing. A social media feed that redraws at 120 fps while the user is reading a static post wastes GPU cycles and display refresh power.
• Inefficient JavaScript: web views and hybrid apps running heavy JavaScript can keep CPU cores at elevated frequencies for extended periods. A poorly optimised ad SDK running continuous layout calculations in the background of an otherwise idle app is a surprisingly common heat source.
• Location polling: apps that request fine GPS location at high frequency (every second instead of every 30 seconds) keep the GPS receiver and associated processing active, adding 0.2 to 0.5 watts continuously.
• Camera processing: social media apps that keep the camera preview active in the background, or that apply real-time ML filters to every preview frame, can push the ISP and NPU power consumption to 1 to 2 watts for a feature the user is not actively using.

Android's battery usage statistics (Settings, Battery, Battery Usage) and iOS's battery health screen give a rough view of which apps are consuming the most energy, but they do not directly show thermal contribution. On Android, the dumpsys thermalservice command on a rooted device shows the current thermal status, mitigation level, and which thermal zones are approaching their limits.

The operating system itself participates in thermal management. Android's thermal HAL (Hardware Abstraction Layer) defines severity levels from NONE through SHUTDOWN, and apps can register as thermal status listeners to reduce their own activity when the phone is warm. Well-behaved apps, including the camera app on most flagships, will reduce frame rate, resolution, or processing complexity when they receive a thermal callback. Poorly behaved apps ignore the callback entirely.

Game engines like Unity and Unreal have adaptive performance APIs (Samsung GameDriver, ARM Adaptive Performance) that query the phone's thermal state and dynamically adjust rendering quality: reducing shadow resolution, lowering draw distance, capping frame rate at 30 fps instead of 60 fps when the phone is thermally constrained. These adaptations are often invisible to the player but can extend comfortable gaming sessions from 10 minutes to 30 or more by keeping the SoC within its sustainable power envelope.

The Future: Where Mobile Thermals Are Heading

The thermal problem is getting harder, not easier. Each generation of mobile SoC packs more transistors into the same die area, and while process shrinks improve efficiency per transistor, the total transistor count grows faster. The A17 Pro has 19 billion transistors. The Snapdragon 8 Gen 3 has about 17 billion. The trend points toward 25 to 30 billion within two generations, and all those transistors generate heat when they switch.

Several developments are changing the picture:

3D chip stacking (used in Apple's M-series Ultra chips and expected in future phone SoCs) places logic dies vertically, increasing the transistor count per unit of footprint area. This concentrates heat in a smaller horizontal area, making spreading harder. Thermal management of stacked dies is one of the hardest problems in semiconductor packaging.

Advanced materials are entering production. Graphene-based thermal interface materials with conductivity above 2000 W/mK in-plane are being qualified for mobile use. Diamond-like carbon coatings on vapour chamber internals can improve the two-phase heat transfer coefficient. These incremental improvements matter when the margins are so tight.

Chiplet architectures (where the SoC is split into multiple smaller dies connected by an interposer or a bridge) distribute the heat sources across a larger area, potentially easing the hot-spot problem. Qualcomm and MediaTek are both exploring chiplet designs for future mobile SoCs, partly for yield reasons but with thermal benefits as a secondary motivation.

Software-driven thermal optimisation is becoming more sophisticated. Machine learning models that predict thermal trajectories and pre-emptively adjust workloads (rather than reacting to measured temperatures) can keep the SoC in a more optimal operating point, squeezing more sustained performance from the same thermal budget. Samsung's Game Booster and Apple's Game Mode both include elements of this predictive approach.

But the hard physics remain. A phone is a sealed enclosure with a small surface area, held against human skin that cannot tolerate temperatures above 44 degrees Celsius. The heat must pass through the same materials, traverse the same thermal resistances, and ultimately be dissipated by the same mechanisms of convection and radiation that governed the first smartphone. As long as phones are thin, sealed, and held in human hands, the warmth you feel will be the signature of an engineering system operating at the ragged edge of its thermal budget.

The next time your phone feels warm, consider what is happening inside: billions of transistors switching at gigahertz frequencies, a radio transmitter pushing milliwatts through an antenna, a lithium-ion cell absorbing and releasing energy, all orchestrated by firmware that monitors temperatures fifty times per second and adjusts voltages within microseconds. The warmth is not a problem. It is the exhaust of computation, managed with a precision that would be invisible if the physics did not insist on reminding you through the palm of your hand.