This is on overview of new type of transistor called GaN (Gallium Nitride) and an evaluation package from GaN system which kindly sent it to me last year(!). I lost touch with them for a while and hence the lateness of this write up.
This is a shot of the eval board (GS-EVB-AUD-xxx1-GS) which is rated at 200 watts/channel into 8 ohm:
Notice the absence of any heatsinks. Here is the accompanied 400 watt switching power supply:
GaN transistors are to the left of the transformer on the bottom with yellow stickers on them. Again, no heatsink due to high efficiency.
Current switching power supplies and amplifiers (Class D) use a type of transistor called MOSFET. These transistors are used in digital systems and are designed to be operated at their extreme operation point (fully conducting or not). This provides high efficiency because the transistor is not kept in less conductive state that generates a lot of losses. What I just described is an ideal transistor. Practical MOSFETs take time to switch and during that time, they generate losses in the form of heat and limit how fast you can switch them. GaN transistors aim to solve this problem as these few slides from GaN systems show:
The capacitors (Cxx) are implicit part of the characteristic of the MOSFET transistor and need to be charged and discharged to lower or raise in the input voltage to a silicon MOSFET. This slows down the part causing that less than ideal ramp to full on. There is also an implicit "body diode" that causes slow reverse recovery which results in more losses. GaN transistors don't have reverse body diodes and lower capacitances:
Here is a measurement showing the full cycle (turn on and off):
The faster switching results in better efficiency as noted:
The graphs are not the same which makes the interpretation hard. But if you draw a line at 50 watts on the left, you see that the GaN based amp has already achieved 90% efficiency whereas the Class D is around 80%.
There is also an important aspect to fidelity. Because of the slow switching time, there needs to be a gap between complimentary pair of MOSFET transistors switching (or else they fight each other resulting in destruction or power loss). This gap directly results in distortion in the output of the class D amplifier. Without mitigation, this would result in very poor performance. The impact here is lowered by using feedback, and a lot of it in the case of Hypex and Purifi amplifiers. High feedback amplifiers are difficult to design and keep stable. GaN transistors switch faster and hence don't create as much distortion to be corrected. You are fixing the problem as the source.
Back to our eval board, this is a very complex device with digital input, processing, etc. There is a mechanism to change level of feedback to show the goodness of GaN without a lot of feedback. All of this makes the board expensive (over $1,000). So I am not sure how many of you want to go and buy it so this is more of a technology overview of what to look for in design from others.
Note that this board was commissioned to be designed for GaN Systems. This is a company that specializes in both low and high voltage GaN transistors. Much of GaN transistor market is focused on very high voltage parts for Electric Vehicles (EV) and such. Those transistors do not bring much to the plate here in audio amplifiers where voltages are below 100 volts. This focus on lower voltage GaN transistors allows these to have the efficiencies I explained above.
GaN Systems Class D Amplifier Measurements
As I noted, this module accepts both types of input, analog and digital. Let's start with analog:
And digital:
Performance here is above average compared to median of all amps I have tested (which lands around 78 dB SINAD). But it not stellar. Speaking with the designer, his focus was to keep the level of feedback low to show the advantage of the GaN transistors not needing much. He was not aiming to produce the best performance possible. Personally I wish he had targeted that given that is what we like to see here.
SNR lands in the same range as our total distortion:
Here is power into 4 ohm:
With a 300 watt power supply, there is not enough juice to drive the amplifier to max power with two channels.
Here is the same using 8 ohm load:
Here is an interesting measurement comparing feedback to no feedback:
We see that the distortion is still under control instead of shooting through the roof.
Most impressive is this measurement:
I don't think I have ever tested a switching amplifier with this level of frequency independence. Even Class AB linear amps struggle to produce such an absolutely clean response. Amplifiers with high amount of feedback run out of gain at higher frequencies and hence distort more. Not this design.
Finally, here is the switching spectrum:
Conclusions
MOSFET transistors have revolutionized computing and switching power designs. It is nice to see a new type become available after decades of refinement of MOSFET. GaN transistors offered by GaN systems show a path to cooler running, more efficient and better performing amplifiers and power supplies. The eval unit unfortunately doesn't show the improved performance due to different target for it. So best to look at commercial implementations that shoot for state of the art implementations.
------------
As always, questions, comments, recommendations, etc. are welcome.
Appreciate any donations using: https://www.audiosciencereview.com/forum/index.php?threads/how-to-support-audio-science-review.8150/
This is a shot of the eval board (GS-EVB-AUD-xxx1-GS) which is rated at 200 watts/channel into 8 ohm:
Notice the absence of any heatsinks. Here is the accompanied 400 watt switching power supply:
GaN transistors are to the left of the transformer on the bottom with yellow stickers on them. Again, no heatsink due to high efficiency.
Current switching power supplies and amplifiers (Class D) use a type of transistor called MOSFET. These transistors are used in digital systems and are designed to be operated at their extreme operation point (fully conducting or not). This provides high efficiency because the transistor is not kept in less conductive state that generates a lot of losses. What I just described is an ideal transistor. Practical MOSFETs take time to switch and during that time, they generate losses in the form of heat and limit how fast you can switch them. GaN transistors aim to solve this problem as these few slides from GaN systems show:
The capacitors (Cxx) are implicit part of the characteristic of the MOSFET transistor and need to be charged and discharged to lower or raise in the input voltage to a silicon MOSFET. This slows down the part causing that less than ideal ramp to full on. There is also an implicit "body diode" that causes slow reverse recovery which results in more losses. GaN transistors don't have reverse body diodes and lower capacitances:
Here is a measurement showing the full cycle (turn on and off):
The faster switching results in better efficiency as noted:
The graphs are not the same which makes the interpretation hard. But if you draw a line at 50 watts on the left, you see that the GaN based amp has already achieved 90% efficiency whereas the Class D is around 80%.
There is also an important aspect to fidelity. Because of the slow switching time, there needs to be a gap between complimentary pair of MOSFET transistors switching (or else they fight each other resulting in destruction or power loss). This gap directly results in distortion in the output of the class D amplifier. Without mitigation, this would result in very poor performance. The impact here is lowered by using feedback, and a lot of it in the case of Hypex and Purifi amplifiers. High feedback amplifiers are difficult to design and keep stable. GaN transistors switch faster and hence don't create as much distortion to be corrected. You are fixing the problem as the source.
Back to our eval board, this is a very complex device with digital input, processing, etc. There is a mechanism to change level of feedback to show the goodness of GaN without a lot of feedback. All of this makes the board expensive (over $1,000). So I am not sure how many of you want to go and buy it so this is more of a technology overview of what to look for in design from others.
Note that this board was commissioned to be designed for GaN Systems. This is a company that specializes in both low and high voltage GaN transistors. Much of GaN transistor market is focused on very high voltage parts for Electric Vehicles (EV) and such. Those transistors do not bring much to the plate here in audio amplifiers where voltages are below 100 volts. This focus on lower voltage GaN transistors allows these to have the efficiencies I explained above.
GaN Systems Class D Amplifier Measurements
As I noted, this module accepts both types of input, analog and digital. Let's start with analog:
And digital:
Performance here is above average compared to median of all amps I have tested (which lands around 78 dB SINAD). But it not stellar. Speaking with the designer, his focus was to keep the level of feedback low to show the advantage of the GaN transistors not needing much. He was not aiming to produce the best performance possible. Personally I wish he had targeted that given that is what we like to see here.
SNR lands in the same range as our total distortion:
Here is power into 4 ohm:
With a 300 watt power supply, there is not enough juice to drive the amplifier to max power with two channels.
Here is the same using 8 ohm load:
Here is an interesting measurement comparing feedback to no feedback:
We see that the distortion is still under control instead of shooting through the roof.
Most impressive is this measurement:
I don't think I have ever tested a switching amplifier with this level of frequency independence. Even Class AB linear amps struggle to produce such an absolutely clean response. Amplifiers with high amount of feedback run out of gain at higher frequencies and hence distort more. Not this design.
Finally, here is the switching spectrum:
Conclusions
MOSFET transistors have revolutionized computing and switching power designs. It is nice to see a new type become available after decades of refinement of MOSFET. GaN transistors offered by GaN systems show a path to cooler running, more efficient and better performing amplifiers and power supplies. The eval unit unfortunately doesn't show the improved performance due to different target for it. So best to look at commercial implementations that shoot for state of the art implementations.
------------
As always, questions, comments, recommendations, etc. are welcome.
Appreciate any donations using: https://www.audiosciencereview.com/forum/index.php?threads/how-to-support-audio-science-review.8150/