Cree readies GaN for low-cost markets
Cree claims its new high power GaN RF transistor sets a new standard in performance and price. Late last month, US-based LED lighting and power device pioneer Cree, unveiled its latest family of high power transistors, specifically targeting wireless telecoms infrastructure markets. Housed in a so-called 'innovative plastic package', the GaN-on-SiC devices include the industry's first 300W plastic-packaged transistor operating at 2.7 GHz. Operating frequencies scale to 3.8 GHz and Psat efficiency is a nice 65%, but critically, device costs look set to hit the low price points the telecoms industry needs to move from industry incumbent silicon LDMOS, to GaN transistors.
"Telecoms is a cost-driven market and the cost of GaN relative to LDMOS has certainly paced its rate of adoption," says Jim Milligan, RF business director at Cree. "We know that GaN offers significant performance advantages over LDMOS, but cost has prohibited its rapid adoption for all but the more niche applications." "But with a cost-focus, we've slashed the price of the packaging to get to a [packaged device] price comparable with LDMOS," he adds. Milligan won't be drawn on exact figures, simply saying costs vary according to a device's operating frequency. But as he highlights, a GaN transistor packaged in a ceramic air cavity package would cost twice as much as the LDMOS equivalent.
Meanwhile the latest plastic-packaged GaN device is nearly half the cost of a traditionally-packaged GaN transistor. So price-wise, Cree is getting close, and as Milligan adds: "It's also a lot harder to impedance match LDMOS at frequencies above 2 GHz, so you don't tend to see silicon in a plastic package at these power levels, and that's how we're fundamentally able to compete." Compatibility counts So how exactly has Cree managed to safely package a high performance GaN transistor in plastic? Again, Milligan will not be drawn on details, but as he explains the packaging is non-standard with Cree researchers 'engineering the die attach process, materials and plastic over-mould system to be compatible with a high power, high frequency GaN transistor die'. The packaging platform is compatible with the various die peripheries used in telecommunications applications with the materials system also being compatible with current plastic overmould practices.
As Milligan points out, Cree will use the same high volume manufacturing processes already used for standard plastic overmould transistors. "But we think our approach is fundamentally different to other GaN vendors whose products in standard plastic overmould packaging are only capable of servicing lower average power applications," he adds. "Instead we have worked with partners to engineer both the mould compound and die-attach processes to be compatible with much higher temperature operation."
Initial adoption is expected in high frequency, high power systems where silicon can't meet performance demands, and Milligan expects GaN will gain significant traction at lower frequencies as pricing continues to fall. "There's going to be new system deployments up in the 3.5 GHz band and we'll likely see GaN there from the start, but it is reasonable to assume that there will be many low frequency applications that will continue to use LDMOS for quite a few years to come," he asserts.
But silicon LDMOS aside, what about up and coming GaN-on-silicon transistors? For example, in February this year, US provider of high performance semiconductors, M/A-COM, bought long-time GaN RF semiconductor developer, Nitronex, later joining forces with UK-based wafer maker IQE to produce four, six and eight inch GaN-on-silicon epiwafers. M/A-COM claims to be in 'active discussions with select companies' to license out its IP, just one recent development that signals the rise of GaN-on-silicon technologies.Milligan concurs GaN-on-silicon will gain market share in RF cell applications, but believes the technology will still remain niche, being used in lower power and frequency applications. "GaN-on-silicon is a tough materials system to manufacture with the epiwafer subject to strain and cracking particularly on the larger wafers that are often touted as a price advantage," he says "In addition to thermal challenges, considerable engineering and yield improvements are required for it to be a viable alternative." "But given the rapid growth and cost reduction we've seen in GaN-on-SiC - we're currently selling millions of transistors per year - and further expected price reductions as we migrate to six inch wafers in the next few years, it's likely this will remain the incumbent GaN technology platform," he adds.
