New trajectories for analog electronics

Posted on

Semiconductor Industry Association (SIA), USA organized a conference today, based on the Decadal plan for semiconductors: new trajectories for analog electronics.

New analog and intelligent sensing systems need to be discovered that can reduce raw data to actionable information, and effective conversion of real-world stimulus to usable entities. David Isaacs, VP, Government Affairs, SIA, introduced the keynote speaker, Jim Wieser.

Five shifts needed!
Jim Wieser, Director of University Research and Technology, Texas Instruments, was the moderator. He said the decadal plan identified five shifts — analog, memory, storage, communications, compute energy, etc. Analog is the interface to the real world. Analog electronics is also driving semiconductors. Seismic shift one is effectively leveraging massive analog data. Analog grand goal is for revolutionary technologies to increase the actionable information with less energy. This will enable efficient and timely (low latency) sensing-to-analog-to-information with a practical reduction ratio of 100,000:1.

New trajectories for analog electronics include sensing and processing, energy efficient functions, such as communications, computing and processing, power conversion and management, as well as bio-inspired models, and holistic co-design. There will be analog bio-inspired ML, THz regime analog, etc. He shared a model of a sensing system approach. Trillions of sensors generate redundant and unused data. Cloud is not the answer. Intelligent sensors are needed to drive local and timely action.

We need the required research and study of holistic solutions, with key apps knowledge and focus on minimal processing to take action. Heterogenous integration is needed to make best use of best technology. There needs to be optimum power management, leverage human systems, and have flexible, secure, and scalable platform and technology, including sensors, memory, and signal representation matched to domain.

David H. Robertson, Senior Technology Director, Analog Devices, moderated the panel discussion. The panelists were Wai Lee, Chief Technologist, Sensing Business, Texas Instruments, Dr. Steven Spurgeon, Staff Scientist, Energy and Environment Directorate, Pacific Northwest National Laboratory, Dr. Mark Rodwell, Doluca Family Endowed Chair in Electrical and Computer Engineering, University of California, Santa Barbara (UCSB), Boris Murmann, Prof. of Electrical Engineering, Stanford University, and Kostas Doris, Fellow, NXP Semiconductors.

Intelligence at the edge
Dr. Steven Spurgeon, Pacific Northwest, said that advanced instrumentation is a catalyst for national scientific innovation and discovery. We must develop new ways to quickly interpret and act on high bandwidth and heterogenous data. We need domain-grounded reduction and inference to unlock the full potential of sensors. There are challenges and opportunities. How can embedded domain knowledge aid in sensing-to-action workflow? How do we harness multi-modal analog data streams? What does co-design look like in specific analytic contexts?

Dr. Mark Rodwell, UCSB, said that transistors for wireless are important. In 5G/6G wireless, terabit aggregate capacities are being served. Currently, there is the exploding demand on wireless networks. We need radar for civilian and military apps. We need to hook up electronic gadgets. CMOS alone won’t do it! Wireless needs low noise, high power and efficiency. Dennard’s scaling laws have been broken. CMOS is optimized for VLSI, but not wireless. CMOS needs help to cover moderate distances.

Wireless needs cheap and high performance technology. It needs app-specific IC wireless technologies, and probably, wireless-optimized CMOS, such as GF 45nm SoI, Intel 22FFL, etc. Heterogenous integration, or, very dense packaging is also needed. There is integration density challenge to be met, along with III-V, heat, production, etc.

Kostas Doris, NXP, noted that autonomous driving needs multiple sensing modalities. Is it simple for Lidar, radar, and camera to be together? There is zoom-in radar evolution. Radar wavelength evolution involves the wider solution space for each step. Much more is needed! We are mapping localizations of radar/sensing needs and 77GHz band limitations. Interference management is also needed.

There are challenges ahead, such as link budget, massive MIMO complexity, data rate explosion, etc. As of now, no technology does it all. To get there, we need multiple functions, in-package integration, multiple nodes, advanced DSP, etc. All of these should be conditioned together to the sensing function.

Wai Lee, Texas Instruments, talked about intelligent sensing and sensor fusion. He used the multi-modal sensing and sensor fusion model. In the next decade, we will see sensor fusion at the edge. There will be A2I, rather than A2D. Besides, there will be compressive sensing with multiple sensing modalities, self-health monitoring of sensors, and low complexity and energy-efficient algorithms for pattern recognition and data security.

There has been compressive sensing and A2I in the past decade. Eg., wearable ECG. There are two research directions to pursue. How can we make these more general purpose by having more intelligence? Also, we need system level optimization to determine the I for A2I. We need flexibility vs. optimization tradeoffs.

Boris Murmann, Stanford University, said that we have communication and inference. Unstructured data has very small information bits per sample. We need to have Interfaces 2.0 that are domain-specific and data-driven architecture design. We need to minimize data conversions, data movement, and memory access. We can combine the strengths of analog and digital for low-energy processing.

Data conversions are expensive, as is memory access. Digital can also become superior. Moving bits is also expensive. The group at Stanford has worked on some ideas, such as pre-distorter spectrum sensing, log gradient imager, RF spectrum sensing, audio feature extraction, etc. The challenge is: how to generalize the amortize the R&D effort across multiple apps? We also need to determine analog block requirements without running huge number of CPU cycles. We need to link the architecture search to relevant low-level circuit specs. We require to educate the next-gen IC designers to embrace higher levels of abstraction.

mmwave packaging challenge
Mark Rodwell, UCSB, said there is also the mmwave packaging problem. There are many kinds of wireless links. We need to make the IC electronics fit. We also need to avoid signal losses and remove heat. Wireless systems in future will have many low-power channels. We will need dense mmwave IC design, low-power, efficient back-end processing, massive low-SNR signals, new digital beamformer designs, and new low-precision array architectures.

We also need advanced interconnects, and not just for VLSI. 5G/6G needs high performance IC interconnects. We need high integration density, high density for efficient power-combining, and high density for efficient multi-finger transistors. Massive MIMO beamforming has many implications. Wai Lee added it is important to get the information from multiple sensors. We need to get bare minimum information for each app.