E2CDA: Type I: Collaborative Research: A Fast 70mV Transistor Technology for Ultra-Low-Energy Computing

包含数十亿晶体管开关的微处理器是个人电脑、手机、回答我们互联网搜索的计算机服务器，以及模拟天气和设计新药和飞机的超级计算机的核心。现代机械是由微处理器控制的；一辆典型的汽车需要50台。经过50年的快速改进，自2000年以来，进展一直停滞不前，主要是因为晶体管在切换时消耗了太多能量。随着晶体管变得更小，更适合芯片，消耗的能量也会增加。电池很快就会耗尽，芯片会变热。减慢切换速度会减少发热，但软件运行速度会很慢。在这个项目中，将研究一种新的晶体管设计。如果成功，这些晶体管的开关能量消耗将减少100倍，从而实现更快、更强大的芯片。挑战是电源电压；将电压降低2：1将开关能量降低4：1。在~1990-2005年间，电压从5伏降至1伏。不幸的是，对于正常的(MOS)晶体管开关，在~0.7伏以下，晶体管的开关变得不完美，晶体管没有完全关断。这种有限的关态泄漏电流增加了能量消耗，因此不可能提供远低于0.7伏的电源。十年前，隧道晶体管被提出，因为即使在低至0.3伏的电源下，隧道晶体管也可以几乎完全关闭。不幸的是，隧道晶体管不能很好地开启，因此使用它们的微处理器将运行缓慢。如果电源降到0.1V，这个限制就会变得更糟。本项目将研究一种新的设计，即三异质结隧道晶体管。这增加了半导体结层，使导通电流增加了100：1。如果成功，即使是0.07V的电源，快速切换微处理器也是可行的，并且消耗的能量只有今天技术的1%。

Microprocessors containing billions of transistor switches are at the heart of PCs, cell phones, computer servers answering our internet searches, and supercomputers modeling the weather and designing new drugs and aircraft. Modern machinery is controlled by microprocessors; a typical car uses 50. After 50 years of rapid improvement, since 2000 progress has stalled, primarily because the transistors consume too much energy when they switch. As transistors are made smaller, more fit on a chip, and the energy consumed increases. The battery is drained quickly and chip becomes hot. Slowing the switching reduces heating, but then the software runs slowly. In this program, a new transistor design will be investigated. If successful, these transistors will consume 100 times less switching energy, allowing faster, more powerful chips. The challenge is the power supply voltage; reducing the voltage by 2:1 reduces the switching energy 4:1. Between ~1990-2005, voltages were reduced from 5 to 1 Volt. Unfortunately, with normal (MOS) transistor switches, below ~0.7 Volts the transistor's switching becomes imperfect, with the transistor not turning completely off. This finite off-state leakage current increases energy consumption, hence it has not been possible to supplies much below 0.7 Volts. Ten years ago, tunnel transistors were proposed, as these can turn off nearly completely even at supplies as low as 0.3 Volts. Unfortunately, tunnel transistors do not turn on well, and microprocessors using them will therefore operate slowly. This limitation becomes much worse if the supply is dropped to 0.1 Volts. This program will research a new design, the triple-heterojunction tunnel transistor. This has added semiconductor junction layers which increased the on-current by as much as 100:1. If successful, rapidly-switching microprocessors will be feasible with even a 0.07V supply, and would consume as little as 1% of the energy of today's technology.