Moore’s Law is getting closer and closer to the end of the line but a recent announcement from Intel pretty much guarantees it will have the legs to carry on until the end of the decade.
Their recently minted 20nm transistors switch on and off 1.5 trillion times per second, making it the world’s fastest transistor, according to Rob Willoner, market analyst in the technology and marketing group at Intel’s Santa Clara, Calif. headquarters.
The increased speed and reduction in size will help provide the necessary computing power to edge us closer to the Star Trekian world of true voice-enabled computing.
Though the transistor will not be on the market for another half-dozen years, the announcement of the limited batch creation is important due to the actual technology used to create them, analysts said.
“Probably the most significant thing is that the transistor is made with technology that exists today,” said Dan Hutcheson president of VLSI Research in San Jose, Calif.
“So now the problem is not a science issue, it is a manufacturing issue… We now know we can make it with conventional technology and that is what is so impressive about it,” he added.
The importance of this announcement is that Intel’s development underpins the future of transistor technology without having to rely on untested (and largely unknown) methods and material to carry on with Moore’s law.
“If those new methods and materials don’t come to fruition, Moore’s law doesn’t stop,” he explained.
“It is more evolutionary than revolutionary, we didn’t have to do anything radically different,” Willoner admitted.
small is good
Since the transistor is much smaller than its predecessors many more can eventually be placed on a chip. Today’s Pentium 4 processors have 42 million transistors on the chip, according to Intel.
Using the 20nm transistors (the measurement is from one side of the transistor gate to the other, with today’s transistors being about 6.5 times wider) Intel will be able to place nearly one billion transistors on each chip. The net result is chips with speeds approaching 20GHz by 2007, according to Intel.
Another big plus according to Willoner is reduced voltage each generation of transistors needs to operate. Reducing required voltage by 15 per cent reduces the power required by 30 per cent he said.
Despite this, a chip with more transistors will obviously require more power regardless if each individual transistor is using less.
Willoner said one of the areas being focused on is the actual gate oxide which acts as an insulator protecting the channel from the gate electrode. If it gets too thin (the 20nm transistor has a 0.8nm gate, pretty much the physical limits of silicon dioxide) current leaks, if it is too thick then you have a slower transistor.
Many companies (Intel included) are looking into substances called High k dielectrics, in essence highly effective insulators, to replace the silicon dioxide.
“We want some material there that will have some properties that will allow us to make it thicker but still keep the transistor as fast as it is today,” Willoner explained.
The High k will be considerably thicker than silicon dioxide and will most likely be the gate oxide used in the 20nm transistor in 2007.
But the steps to bring this new transistor into production will not be easy, analysts said.
diffucult road ahead
“They are showing off a bit…but there is no smoke and mirrors here, they have achieved something [and I] believe they will bring it to market,” said Richard Doherty, director of research at the Envisioneering Group in Seaford, N.Y.
But the next step will be to make the transistors more neighbourly so they work well in close proximity, he explained.
“You are also going to have metal interconnects, I don’t want to rain on Intel’s parade but for all we know the wires leading into the transistors might be thicker than the transistor itself,” he said.
So the interconnects between the transistors will be one of the next challenges, he said.
Regardless he agreed that the High k dielectrics will be the wave of the future even though there are potential problems that could arise.
“[High k which works well] at 1GHz may start to look like a solid chunk of copper at 10GHz because the conductivity of these dielectrics can change with frequency,” he explained.
So there is still a lot of work to be done.
“It is not just Moore’s law anymore, it is can you comercialize it,” Doherty said.
Willoner agreed. “It is a big difference between something that you can show, which is what we are doing here, and something that you can produce at extremely high volume with the reliability that is required with it.”
“We expect to push Moore’s law beyond 2010, we are nine years away from that so we have a lot of time to work between now and then to be able to confidently say we will be able to do that…We see no major hurdles to pushing Moore’s law through the end of this decade,” he added.
“We are so excited about this we just wanted to tell the world what we had achieved,” he joked as he concluded.