I. INTRODUCTION

AlGaN/GaN HEMT has a high saturation velocity thanks to the 2-DEG channel formed by polarization and is mainly used for RF power amplification ^[1,2,3]. However, self-heating leads to an increase in channel temperature in high-power operation conditions, which affects the device's performance and reliability. Therefore, channel thermal resistance is considered to be one of the most important parameters for robust reliability. SiC substrate possesses higher thermal conductivity than Si and has been preferably employed in GaN RF applications to mitigate the thermal effect ^[4,5,6].

Conventional RF power AlGaN/GaN HEMT devices use carbon or iron-doped GaN buffers to reduce the conductivity and minimize leakage current, thereby increasing breakdown voltage ^[7,8]. However, during RF operation, the Mg doping can act as an acceptor trap, causing DC-RF dispersion, widely known in GaN device technology ^[9,10]. In addition, the distance between the substrate and the channel is longer due to the thicker buffer, which is disadvantageous in releasing heat generated in the channel. To solve this problem, a `thin buffer' structure has recently been adopted ^[11]. The thin buffer structure uses a very thin unintentionally doped GaN (UID) buffer with a back barrier to reduce the leakage current.

In this paper, the channel thermal resistance of the Fe-doped thick GaN buffer and the thin buffer AlGaN/GaN HEMT devices are extracted and compared through an electrical method. The pulsed I-V measurement experiment was conducted with increasing the channel temperature in two ways: an increase in the channel temperature by the dissipated power P${}_{\rm D}$ (self-heating) and an increase in the channel temperature by the chuck temperature (external-heating). By matching I${}_{\rm D.SAT}$ which was decreased by the increased channel temperature, we were able to extract the channel thermal resistance.

To prevent the electron trapping by high drain bias, the measurement was carried out with a low voltage of 5V or less. The gate current was so small ($<1$ uA/mm) that heating by gate current was neglected.

II. DEVICE STRUCTURE

Fig. 1 shows the cross-sectional structure of the AlGaN/GaN HEMT with a different buffer thickness. The devices measured in this work have a gate length and width of 0.15 $\mu$m and 2*50 $\mu$m, respectively. An AlN back barrier was employed under the GaN buffer to reduce the leakage current. The thick buffer structure consisted of AlN nucleation, 2 $\mu$m Fe-doped GaN buffer, 20 nm Al${}_{0.255}$Ga${}_{0.745}$N barrier (Fig. 1(a)). In the thin buffer device, AlN nucleation layer was grown on the SiC substrate followed by a 0.25 $\mu$m unintentionally doped (UID) GaN layer. Then the 1nm AlN insertion layer, 10nm of AlGaN barrier layer with 30 % Al content and a 2 nm GaN cap layer were followed (Fig. 1(b)).

Fig. 1. Schematic of the HEMT device structure at (a) thick-buffer HEMT, (b) thin buffer HEMT.

III. CHANNEL THERMAL RESISTANCE EXTRACTION

DC and pulsed I-V measurement were performed by using Keithley's 4200-SCS parameter analyzer with pulse measurement unit(PMU) and DC output characteristics are shown in Fig. 2 for thick buffer (Fig. 2(a)) and thin buffer devices (Fig. 2(b)). In typical pulsed I-V measurement, the quiescent bias is applied during the certain period time, then the short pulse bias is applied to the drain or gate, or both electrodes and the current is measured during the pulse condition. The quiescent voltages on the gate were set to V${}_{G.Q}$ = 0 V for the thick buffer device. For the thin buffer devices, V${}_{G.Q}$ was set to 1 V to allow sufficient self-heating by increasing the channel current during the quiescent condition. The pulse-width and duty cycle were set to 1 $\mu$s and 0.1%, respectively, ensuring no heating occurred during measurement.

Fig. 2. Output I${}_{D}$V${}_{D}$ characteristics of a) thick-buffer HEMT b) thin buffer HEMT.

1. Self-heating extraction

The channel thermal resistance was extracted from the pulsed I-V measurement by matching the reduced drain saturation current after increasing the channel temperature through self-heating and external-heating ^[12,13].

In general, pulsed I-V can measure the current without the self-heating effect by using a short measurement voltage pulse. In this case, the drain quiescent voltage (V${}_{\rm D.Q}$) is set to 0 V, no current flows through the channel, resulting in no power dissipation.

However, the dissipated power (P${}_{\rm D} =$ I${}_{\rm D.Q} *$ V${}_{\rm D.Q}$) in a semiconductor channel is emitted as heat, leading to an increase in the channel temperature if V${}_{\rm D.Q}=$ 5 V is applied with V${}_{\rm G.Q} =$ 0 or 1 V. Fig. 3 illustrates the reduction of I${}_{\rm D.heated}$ due to the self-heating generated during the quiescent period.

When the drain quiescent voltage increases from 0 V with the 2 DEG channel formed which is the case of V${}_{\rm G} =$ 0 or 1 V, the channel temperature elevates with more P${}_{\rm D}$. Upon applying a very short pulse of V${}_{\rm D.M} = 5$ V to measure the current at this elevated temperature, the current will be lower than the pulse current measured at V${}_{\rm D.Q} =$ 0 V, resulting from the self-heating induced during the quiescent period. By progressively increasing the V${}_{\rm D.Q}$ and extracting I${}_{\rm D.heated}$, a graph can be plotted that shows the pulse current as a function of P${}_{\rm D}$ in the quiescent state.

Figs. 4(a) and 4(b) show the extracted I${}_{\rm D.heated}$ as a function of P${}_{\rm D}$ for the thick buffer device and the thin buffer device, respectively. In both cases, as P${}_{\rm D}$ increases, the channel temperature rises, resulting in a decrease in I${}_{\rm D.heated}$.

Fig. 3. The strategy of puled I-V measurement with or without self-heating effect.

Fig. 4. I${}_{\rm D.heated}$ with different P${}_{\rm D}$. measured from (a) thick buffer and (b) thin buffer devices.

2. External-heating extraction

External-heating extraction was performed by increasing the base-plate temperature to increase the temperature of the channel and extracting the corresponding I${}_{\rm D.heated}$ reduction. When the temperature of the base plate was increased under the condition where the V${}_{\rm D.Q} = 0$ V, without self-heating, the channel temperature rose due to the external heat. In this case, it can be assumed that the base plate and channel temperatures are equivalent. Under these conditions, increasing the base-plate temperature will further raise the channel temperature, leading to a reduction the I${}_{\rm D.heated}$. By progressively increasing the base-plate temperature and extracting the corresponding I${}_{\rm D.heated}$, a graph can be plotted to show I${}_{\rm D.heated}$ as a function of base-plate temperature. As shown in Figs. 5(a) and 5(b), as the temperature of the chuck increases, the temperature of the channel increases, and I${}_{\rm D.SAT}$ decreases.

Fig. 5. Extracted I${}_{\rm D.heated}$ with different chuck temperatures at a) thick buffer device b) thin buffer device.

3. Channel thermal resistance extraction

From the two measurements under self- and external heating, the relations of I$_{\rm D.heated}$ by P$_{\rm D}$ and base plate temperature were obtained. By matching the P$_{\rm D}$ and base-plate temperature at which the same current was flowing in both graphs, a relation between P$_{\rm D}$ and base-plate temperature, Fig. 6 was obtained.

Fig. 6 presents the R${}_{\rm TH}$ of thick buffer(Fig. 6(a)) and thin buffer devices (Fig. 6(b)), determined by extracting the slope of the trend lines. The extracted R${}_{\rm TH}$ values are $38.794^\circ$C mm/W for the thick buffer device and $23.716^\circ$C mm/W for the thin buffer device. The lower thermal resistance of the thin buffer device compared to the thick-buffer device (approximately 39% lower) was attributed to the more efficient heat dissipation through the high thermal conductivity of the SiC substrate, which acts as a heat sink and facilitates easier heat dissipation due to the closer proximity between the channel and substrate.

Fig. 6. Estimated channel temperature as a function of P${}_{\rm D}$ based on data from Figs. 4 and 5 at a) thick buffer device and b) thin buffer devices.}

IV. CONCLUSIONS

In this paper, the thermal resistance of GaN devices with thick (2 $\mu$m) and thin (0.25 $\mu$m) buffers was evaluated. The analysis of thermal characteristics under self-heating and external-heating conditions indicated that the thermal resistance of the thin buffer GaN device ($23.7^\circ$C mm/W) was lower compared to that of the thick-buffer GaN device ($38.9^\circ$C mm/W). This difference is attributed to the more efficient heat release through the SiC substrate in the thin buffer device. These results suggest that the thin buffer GaN structure can be a strong candidate offering potential advantages for high-power amplification, owing to its improved thermal performance.