15-03-2023 | By Robin Mitchell
Recently, semiconductor manufacturers like Intel have been facing significant challenges in shrinking transistors to the 1nm node due to quantum effects like quantum tunnelling, which violates the laws of physics. To overcome this, companies are investing heavily in material sciences, circuit design, and patterning technologies to improve insulative materials and reduce high-speed processors' operational currents and energy consumption. With Intel announcing the tape out of its latest 20A and 18A processes, the industry is one step closer to overcoming these challenges and reaching the 1nm node. What challenges have semiconductor manufacturers faced as they approach the 1nm mark, what technology will Intel be leveraging, and what could this mean for future semiconductors?
Ever since the first integrated circuits, semiconductor foundries have had to face numerous challenges regarding circuit design, material sciences, and even the laws of physics. By shrinking down transistors, engineers can either increase the number of dies from a single wafer, increase the number of circuits on each die, or a mixture of both. At the same time, shrinking transistors has also had the effect of reducing the energy consumption of each transistor, enabling lower-powered devices. This results from the use of smaller channels and gates, which reduce the current needed to change the state of transistors. In addition to lower operational currents, smaller transistors are also quicker to switch, thereby enabling the development of high-speed processors, which are now beginning to exceed 6GHz.
However, recent semiconductor industry developments present engineers with challenges that are becoming increasingly more challenging to solve. For example, as process nodes move towards the individual nanometres, quantum effects such as quantum tunnelling become far more apparent, and this can have a serious impact on the performance of devices. Unlike typical current leakage, quantum tunnelling cannot be solved simply by using a more insulative material, as quantum tunnelling occurs irrespective of the barrier between two points. Thus, engineers have had to find extraordinarily unusual and innovative solutions to try and get around such phenomena.
Producing transistors at the nanometer scale also introduces challenges with patterning. Older semiconductors had feature sizes large enough that visible and UV light could be passed through a mask which is then shrunk down using a series of lenses onto the semiconductor target. However, devices on the nanometer scale are far smaller than the wavelength of light commonly used in the semiconductor industry, meaning that extreme UV has to be used (which still has a wavelength much larger than the features found on modern devices) in conjunction with complex mask techniques. To put into perspective just how complex this process is, only one company in the entire world, ASML, has the capability to produce EUV machines.
Recently, Intel announced that it had finalised its plans for 20A and 18A devices, with some suspecting that device prototypes have already been taped out, ready for testing. According to Intel, their 20A and 18A are six months ahead of schedule, which would put commercial devices ready for Q4 2024, with the 20A process likely to be ready before the 18A. Additionally, Intel also noted that their 20A and 18A processes would benefit from RibbonFET and PowerVia technologies, leaving FinFET technology behind.
It is expected that the 20A process will provide engineers with a 15% improvement per watt, while the 18A process will provide a 10% improvement in performance per watt, but it is unclear if that is a 10% improvement over the 20A node. Intel also announced that its Intel 4 node process (4nm) is still due to be launched, but at this current rate, it would seem that Intel will be releasing four nodes in the span of two years.
If Intel can deliver on its 18A process node, it indicates a positive trend for the semiconductor industry, as Moore’s law will still be achievable until the end of 2030, meaning that engineers can expect devices to increase transistor densities. At the same time, the ability to break the 5GHz barrier without requiring extensive cooling techniques also indicates that single-thread speeds will increase, further providing massive performance boosts.
Additionally, introducing 20A and 18A technologies will significantly benefit the wearables and mobile market, which relies heavily on low-energy devices. With the new technologies being offered by Intel, it wouldn’t be surprising if developments in VR and AR began to accelerate while reducing the overall cost of such systems. It may even become possible for wearable devices to contain all computational needs for processing VR data in real-time, something which high-end graphics cards are currently needed for.
Finally, it is also likely that new features such as AI and ML will routinely be integrated into commercial products using these new technologies offered by Intel. One such possibility could be the development of real-time hardware security systems that operate simultaneously with a processor’s execution, actively looking for suspicious activity while having zero performance impact. Such computing systems could even explore the use of large-scale FPGAs for providing reconfigurable hardware, allowing customers for the first time in history to apply hardware updates.