Key points of high-frequency packaging structure for discrete devices

High-Frequency Discrete Packaging: The Parasitics That Kill Your Signal Before It Leaves the Board

At low frequencies, a discrete transistor is just a switch or an amplifier. At high frequencies — say, anything above 100 MHz or into the gigahertz range — that same device becomes a complex network of inductors, capacitors, and resistors. The package is no longer a container; it is the circuit.

Every millimeter of lead length is an inductor. Every square millimeter of die area is a capacitor. The mold compound is a lossy dielectric that eats your gain. If you are designing for RF power amplifiers, switch-mode power supplies operating at high speed, or fast logic, the package parasitics are the bottleneck. You can have the perfect silicon die, but if the package adds 2 nanohenries of series inductance, your device is useless at 2 GHz.

High-frequency packaging is about minimizing those parasitics and controlling the impedance until the signal hits the PCB trace.

The Inductance Problem: Why Leads Are the Enemy

The Physical Limit of Wire Bonds

The wire bond is the Achilles heel of high-frequency discrete packaging. A standard gold wire bond is 25 micrometers in diameter and 2 to 3 millimeters long. At DC, that is a short circuit. At 1 GHz, that wire is an inductor with roughly 1 to 2 nanohenries of inductance.

Inductance kills switching speed. When you try to turn a MOSFET off, the current through the inductance cannot stop instantly. The voltage spikes (V = L * di/dt), ringing occurs, and you get electromagnetic interference that fails your FCC tests.

For high-frequency operation, you must eliminate the wire bond from the main current path. This is why "clip" packages or "flip-chip" style discrete packages exist. Instead of a wire loop, a flat copper strip connects the die to the lead frame. The loop area is near zero, which drops the inductance to less than 0.2 nanohenries.

If you are stuck with a standard wire-bonded package for cost reasons, keep the bond wires as short as possible. A 1 millimeter bond wire has half the inductance of a 2 millimeter wire. But do not expect magic. Wire bonds are fundamentally low-frequency structures.

Lead Frame Geometry and Skin Effect

At high frequencies, current does not flow through the whole cross-section of a conductor. It flows on the surface. This is the skin effect. For a copper lead frame at 500 MHz, the skin depth is about 3 micrometers.

This means the center of your lead frame is dead weight. It adds mass and thermal capacity but carries almost no current. The effective resistance of the lead frame increases with frequency, causing I-squared-R losses that heat up the package.

The fix is surface plating. Silver or gold plating on the lead frame reduces the surface resistance because silver has higher conductivity than copper at high frequencies. A silver-plated lead frame can reduce high-frequency conduction losses by 15 to 20 percent compared to bare copper.

Capacitance and the Miller Effect

The Gate-Drain Capacitance Trap

Inside the package, the die has internal capacitances. The most critical one for high-frequency switching is the gate-drain capacitance (Miller capacitance). In a standard plastic package, the lead frame and the mold compound add parasitic capacitance to this node.

This extra capacitance couples the output signal back to the input. When the drain voltage swings high, it kicks the gate voltage up through this capacitance, potentially turning the device back on when you are trying to turn it off. This is parasitic turn-on, and it destroys efficiency in Class D amplifiers and high-speed converters.

Package design minimizes this by shielding the gate lead from the drain lead. In RF packages, the gate pin is often placed between two ground pins, or the lead frame is designed so the gate-to-drain distance is maximized within the package outline.

Mold Compound Dielectric Loss

The epoxy mold compound is not a perfect insulator. It has a dissipation factor (loss tangent). At high frequencies, the electric field in the mold oscillates, and the polar molecules in the epoxy try to follow the field. They lag behind, generating heat.

Standard epoxy has a loss tangent of 0.02 to 0.03. That sounds low, but at 2 GHz, that loss eats your signal gain. For RF power devices, the mold compound must be a low-loss ceramic-filled epoxy or a pure polymer with a loss tangent below 0.005.

If you ignore this, your power amplifier will have 2 dB less gain than the datasheet promises. That missing gain is dissipating as heat inside the plastic package, cooking the die from the outside in.

Thermal Management at High Frequency

Thermal Impedance vs. Thermal Resistance

At DC, we talk about thermal resistance (Rth). At high frequencies, we must talk about thermal impedance (Zth). The die heats up in pulses, not a steady state. The package has thermal capacitance — it takes time for heat to travel from the junction to the case.

If the pulse repetition rate matches the thermal time constant of the package, the junction temperature spikes much higher than the average power calculation suggests. A device rated for 5 watts average might burn out at 2 watts average if it is switching at 1 MHz with a 50 percent duty cycle.

The package must have low thermal impedance at the operating frequency. This means thinning the die attach layer. Sintered silver die attach is essential here because it has high thermal conductivity and low thermal mass compared to epoxy. It moves heat out of the die fast enough to keep up with the pulse train.

Via Inductance Under the Pad

For high-frequency discrete devices, the thermal vias under the exposed pad are also part of the RF ground return path. A via is an inductor. A standard 0.3 mm via has about 0.5 to 0.7 nanohenries of inductance.

If you have 12 vias under the pad, and they are spaced out, the total inductance of the ground return path can be 5 to 10 nanohenries. At 500 MHz, that is a significant impedance. It degrades the bypassing of the supply rail and allows RF noise to escape onto the power bus.

The solution is via fencing. Place a ring of vias close to the pad edge, connected to the ground plane. This creates a low-inductance shield around the thermal path. The vias must be solder-filled or copper-filled to minimize inductance and maximize thermal conductivity.

Package Types for High Frequency

The SOT-323 and SOT-363 Limits

SOT-323 (SC-70) and SOT-363 are popular for small signal RF transistors. They are tiny, which is good for low inductance. But they have a fatal flaw: the thermal pad is small, and the lead frame is thin.

At power levels above 200 milliwatts, the SOT-323 hits thermal limits fast. The thin lead frame cannot spread the heat, and the small pad cannot conduct it to the board. For high-power RF (like LDMOS or GaN devices), you need to move to SOT-223, DPAK, or QFN packages.

SOT-223 is the workhorse for RF power up to 1 watt. It has a large enough tab for heatsinking and a lead frame robust enough to handle the current. The inductance is higher than SOT-323, but manageable up to about 1 GHz.

QFN and DFN for GHz Operation

For true gigahertz operation, QFN (Quad Flat No-leads) and DFN (Dual Flat No-leads) are the only choices. The "no-leads" part is critical. There are no wire bonds sticking up. The connections are flat pads on the bottom.

This reduces the series inductance to near zero. The signal path from die to PCB is a direct copper-to-copper transition. The package becomes a transmission line structure rather than a lumped element circuit.

The downside is assembly. You need solder paste on the bottom pads, which requires precise stencil alignment. And rework is nearly impossible because you cannot grab the part with tweezers without heating the whole board. But for performance, there is no substitute.

PCB Interaction: The Package Does Not End at the Board

The Launch Transition

The most critical part of a high-frequency design is not the package and not the PCB. It is the transition between them.

If the pad on the package is 0.5 mm wide and the trace on the PCB is 2.0 mm wide, you have an impedance mismatch. The signal reflects at that junction. You get standing waves, voltage peaks, and erratic behavior.

The PCB pad must taper smoothly from the package size to the trace width. A linear taper over 2 to 3 millimeters works well. This creates a broadband impedance match that minimizes reflections up to several gigahertz.

Ground Plane Continuity

High-frequency discrete devices need a solid ground plane directly under the package. If the ground plane has slots or cuts near the device, the return current has to flow around the slot. This increases the loop area, which increases inductance and radiation.

Keep the ground plane solid under the device and for at least 3 millimeters around it. Stitch the top and bottom ground planes with vias around the perimeter of the package. This creates a Faraday cage effect that contains the RF energy and prevents it from coupling into nearby circuits.

If you ignore ground plane continuity, your "high-frequency" discrete device will radiate more noise than a cheap wire. The package can only do its job if the board cooperates.